A store leader's first screen of the day
Before the doors open, the leader needs the numbers, the plan from the leadership briefing, and who needs recognition. The dashboard puts all of it on one scroll.
Austin, Texas
I'm Tyler Hooks, a senior design technologist at H-E-B Digital. I turn emerging technology into things people can test: prototypes that ship to hundreds of stores, AI tools that change how a design team works, and testing that fixes the prototype between research sessions. Before that, I built component libraries at Dell and Disney.
A store-partner tool prototyped from the ground up, now live in 300+ H-E-B stores.
H-E-B · Prototyping · ShippedProject →Bringing Copilot and Gemini Enterprise to a design organization, plus the doc site and Instruction Builder that made them usable.
H-E-B · AI adoption · Internal toolsCase study →Dana asked teams to prep for this weekend's storm: order up on key items, adjust schedules for the lift, and trim overtime today. Check training completions before Friday.
The briefing, payroll, team development, and notes in one app, built and tested in 15 stores.
Turning scattered store tools into one app, shaped by the leaders who’d use it.
H-E-B · RITE testing · Design researchProject →My pappy done die befo' I 'member him. "One day Massa Chapman call all us to de front gallery. Us didn't know what gwine to happen, 'cause it not ord'nary to git called from de work. Him ring de bell and dat am sho' 'nough de liberty bell,
'cause him read from de long paper and say, 'You is slaves no more. You is free.'"
Cited AI search over the Texas WPA Slave Narratives from the Library of Congress.
RAG systemProject →The retrieval step is the design problem. Show what was found, how well it matched, and where it came from — before a single generated word appears.
About this project →A side project for learning RAG end to end on Google Cloud: Vertex AI RAG Engine finds the passages, Gemini answers from only those, and a test set I'm writing keeps it honest. These narratives make a good stress test, because the transcripts spell speech phonetically, so search has to work by meaning instead of matching words.
They're first-person accounts recorded by the Federal Writers' Project in 1936–38 from formerly enslaved people in Texas, and they include painful history and period language. Every answer cites the passages it was built from and says so when they don't cover your question. Naming a narrator gets the best results.
Pick a question, or type your own.▍
A store-partner tool prototyped from the ground up, now live in 300+ H-E-B stores.
Every day, H-E-B store partners complete in-store walks and inventory requirements to keep shelves stocked and counts accurate. A guided task management system walks them through that work. By the time of the redesign, partners were moving between four separate guided walks (store walk, shelf replenishment, shelf inventory, and perpetual adjustment), each with its own flow.
I prototyped the system as a React web app that guides partners through walks and inventory tasks step by step. That prototype became the app now live in 300+ H-E-B stores.
I led the high-fidelity prototype for a redesigned partner experience that replaces all four walks with a single guided flow. I built it in Next.js and TypeScript with an Airtable backend, and deployed it on Kubernetes so it held up through testing.
We ran one-hour, in-person sessions in Austin, Round Rock, and San Antonio on the Zebra devices partners carry every day, scoring each task on whether partners could complete it and understood the new flow.
Screens recreated to protect proprietary work. Full walkthrough available on request →
The system is live in 300+ H-E-B stores. In testing, partners completed 76% of tasks in the redesigned flow outright, and 84% including tasks they mostly understood. That validated the consolidation before engineering handoff, and the 12% of unclear tasks showed exactly where to iterate next.
Building a tool and redesigning it after partners have lived with it are different jobs. The first time, I was proving the idea could work. The second time, I was asking people to give up habits they'd built around my own app, so testing on their real devices, in their real stores, mattered even more.
Bringing Copilot and Gemini Enterprise to a design organization, plus the doc site and Instruction Builder that made them usable.
[TBD]
[TBD]
[TBD]
[TBD]
[TBD]
[TBD]
Turning scattered store tools into one app, shaped by the leaders who’d use it.
Dana asked teams to prep for this weekend's storm: order up on key items, adjust schedules for the lift, and trim overtime today. Check training completions before Friday.
Training for seasonal team members starting in November, focused on guest service and product knowledge. One action item still needs a date.
Store leaders run operations, lead partners, and take care of customers, but their work is spread across apps that don’t connect. Every hour spent at a desk switching between apps and digging through reports is an hour away from the floor. H-E-B wanted to know whether one app could replace them, so we took it to leaders to find out.
I built a high-fidelity prototype in Figma Make covering daily briefings and dashboards, note tracking, partner development, and payroll, so leaders could react to their actual job instead of a concept.
I embedded with Design Research across 15 stores in 4 regions, testing the prototype with leaders and running 26 in-depth partner interviews to learn how their days actually run.
We used rapid iterative testing (RITE), updating the prototype between sessions as stores showed us what they needed. In stores already piloting Gemini, we asked leaders how they were using AI day to day. The briefing page started as a simple summary with follow-up tasks, until we kept seeing leaders jot follow-ups on their phones, on scraps of paper, even on their hands. That became its own notes page with AI summarization. Building in Figma Make, and later GitHub Copilot, kept the prototype rough enough to change overnight and kept every session on one question: does this make a leader’s day easier?
Screens recreated to protect proprietary work. Full walkthrough available on request →
Dana asked teams to prep for this weekend's storm: order up on key items, adjust schedules for the lift, and trim overtime today. Check training completions before Friday.
Before the doors open, the leader needs the numbers, the plan from the leadership briefing, and who needs recognition. The dashboard puts all of it on one scroll.
Sales, shrink, and hospitality sit in a swipeable row, one tap from the payroll forecast.
Managers who missed the leadership briefing still start the day aligned. The summary is generated from the meeting notes.
Each item from the briefing becomes a task with a person and a priority, tracked back to where it came from.
Work anniversaries and team wins sit on the same screen as the numbers, so they get said out loud.
Dana asked teams to prep for this weekend's storm: order up on key items, adjust schedules for the lift, and trim overtime today. Check training completions before Friday.
Training for seasonal team members starting in November, focused on guest service and product knowledge. One action item still needs a date.
The first build treated notes like the dashboard's briefing: an AI summary of the meeting and a few follow-ups. Leaders still jotted the rest on their phones, on scraps of paper, even on their hands, and anything they needed to share went out by email that nobody on the floor was checking.
Drag the handle on the phone to compare v1 and v2.
An AI summary and key topics sit on top, so a partner catching up between tasks gets the point in seconds. The leader's full note stays right below it.
Instead of an email nobody on the floor reads, the note goes to the people who need it, and the leader can see who has acknowledged it.
Follow-ups become action items with an owner and a priority, added right from the note instead of written on a hand.
Questions and answers live on the note itself, so the context doesn't scatter across texts and inboxes.
By the end of the study, the prototype had grown from a fixed artifact built before the first session into a research tool that changed with the evidence while the study’s goals stayed the same. Every round of sessions sharpened the details: store-specific terms, who in each role and department can see what, and generated text that sounds like real floor talk. The result was a store leader experience shaped by real workdays in 15 stores, not by assumptions made before the study.
RITE testing only works if you can change the prototype before the next session. Complex updates used to take too long, so feedback went into a backlog for after the study. AI-assisted prototyping closed that gap without skipping any research steps.
Speed still needed a filter. Reacting to one session risks chasing one person’s opinion, so every change went through synthesis with Design Research and was sorted into three buckets: trust and usability blockers, high-value hypotheses, and longer-term insights. Speed also raised the bar for quality. One wrong store number, or a line no store leader would actually say, could cost a participant’s trust, so I checked every AI-generated update by hand before it went back in front of a leader. Clean design files and prompts that spelled out what should change, and what had to stay the same, made the output reliable.
The goal was never the most edits. It was the right ones: changes that made the remaining sessions teach us more.
This was also the first time design technology was embedded this deeply in research. We joined synthesis, presented readouts, and ran live demos, which showed how much a design technologist adds while the problem is still being figured out.
Cited AI search over the Texas WPA Slave Narratives from the Library of Congress.
"The end of the war, it come jus' like that--like you snap your fingers." "How did you know the end of the war had come?" asked the interviewer. "How did we know it! Hallelujah broke out--
Him ring de bell and dat am sho' 'nough de liberty bell, 'cause him read from de long paper and say, 'You is slaves no more. You is free.'
About 300 first-person interviews with formerly enslaved Texans, recorded by the Federal Writers' Project in 1936–38. They include painful history and period language. Most interviewers were white Southerners working during Jim Crow, which shaped what people felt safe saying, so the system reports individual accounts and never generalizes one person's experience to everyone.
The narratives have been public for years, but they're underused. Interviewers transcribed speech phonetically ("massa," "fotches," "patterrollers"), so keyword search misses passages that are plainly about what you asked. Those spellings were the interviewers' choices, not the speakers'. Different interviewers spelled the same words differently, so the record carries their habits as much as the speakers' voices. The question this project tests: can semantic retrieval bridge that gap, and can the answer stay honest about where it came from? The first test run suggests the second part holds. The first is more complicated.
I used Project Gutenberg's human-proofread text instead of the Library of Congress page scans, which would have needed OCR. LOC stays the cross-reference. A Python script splits the volumes on the six-digit interview number stamped on each typescript, a reliable boundary that ties every file back to the LOC original. One file per narrative, named for its part, interview number, and speaker, means every retrieved chunk carries its own citation. A spot-check caught transcriber back-matter appended to the last narrative of a volume; I fixed the splitter rather than the file.
The corpus runs on Vertex AI RAG Engine in Serverless mode with RAG-managed vector search. At a few thousand chunks, any vector store performs about the same, so the deciding factors were no infrastructure to run and no idle hourly billing. I chunked at 512 tokens with 128 overlap instead of the 1024/256 default: many narratives are 700–2,000 words, and 1024-token chunks would blur several episodes into one embedding. Setup had its own obstacles. The project-level deployment mode defaulted to Spanner, which is capacity-restricted for new projects in us-central1, and the console had no way to change it, so I switched to Serverless through the REST API.
A prebuilt agent in Agent Studio worked, but it pulled in outside sources. The demo makes two calls instead: retrieve the top five passages, then ask Gemini to answer from those five and nothing else. The passages on screen are exactly what the answer was built from, so the answer can't cite anything the visitor can't read. The prompt also tells the model to cite every claim, decline when the passages don't cover a question, correct false premises, report conflicting accounts, and keep speakers' own wording.
Building the pipeline was the easy part. Knowing whether it works is harder, and that's where most of my time is going now.
To start, I had AI draft 78 test questions from 12 narratives I'd read: questions about one specific narrative, at increasing difficulty; questions about themes that run across many; the same question in plain English and in phonetic spelling; questions that invite a sweeping generalization; and questions the system should refuse. The second figure shows what each one is testing.
Those questions are a first draft, and I'm treating them that way. AI-written questions tend to borrow the source's wording, which makes them easier than they look, and a few already turned out to misread the passage they were based on. I'm reviewing them as I read through the narratives myself.
How did people learn they were free?
"I don't know when my mother was set free. My husband's marster's name was King. He was from Savannah, Georgia, but at the time was...
"The end of the war, it come jus' like that--like you snap your fingers." "How did you know the end of the war had come?" asked the interviewer. "How did we know it! Hallelujah broke out--
Him ring de bell and dat am sho' 'nough de liberty bell, 'cause him read from de long paper and say, 'You is slaves no more. You is free.'
Keyword search catches two of these because they happen to say "free." It misses Felix Haywood, whose answer is "Hallelujah broke out." That passage shares no words with the question, and it ranked second by meaning.
"I don't know when my mother was set free. My husband's marster's name was King. He was from Savannah, Georgia, but at the time was...
MATCHED ON: set free
"The end of the war, it come jus' like that--like you snap your fingers." "How did you know the end of the war had come?" asked the interviewer. "How did we know it! Hallelujah broke out--
MATCHED ON: hallelujah broke out ≈ learned they were free
Him ring de bell and dat am sho' 'nough de liberty bell, 'cause him read from de long paper and say, 'You is slaves no more. You is free.'
MATCHED ON: de long paper ≈ a proclamation
All three came back from one plain-English question. The numbers are cosine distances from the live corpus; lower is closer.
What happened when patrols caught enslaved people off the plantation without a pass?
What did the patterrollers do when they cotched folks without a pass?
Neither phrasing reached Toby Jones. But both landed on the same narrators, and the plain question about "patrols" pulled passages that spell it "patter rolls" and "patters." Meaning crossed the spelling gap, just toward different narrators than I expected.
Horses frightened and running away with a carriage
De hosses skeert and run off with de coach
Spelled the way the transcript spells it, the question found Lewis Jones at #1. In plain English it missed him; its top result was a different runaway, Albert Hill's balky horse. Relevant, but every plain result scored weaker than the phonetic version's weakest (0.457 vs 0.428).
Each narrative mostly sticks to one spelling of "master," and they disagree with each other. That points to the person transcribing, not the person speaking. Search has to treat all four as one word. Sample: the 18 narratives read so far; two patroller spellings come from round 1 search results.
Source · Albert Hill · #420007“I heered to tie de rope on de balky hosses tail and run it ’twixt he legs and tie to de shaft. I done dat and puts some cuckleburrs on de rope, too.”
Who tie de rope on de balky hosses tail and puts cuckleburrs on de rope?
Who tie the rope on the balky horses tail and puts cockleburs on the rope?
Who describes a trick to get a stalled animal moving that backfired?
The plain version may differ from the phonetic one only in respelled words (highlighted), so any difference in results comes from spelling. Round 2 sorts each concept by gap type: phonetic spelling, period vocabulary, or grammar.
Round 1's pairs didn't always describe the same details, so a difference in results couldn't be pinned on spelling alone. Round 2 fixes the wording first. Concepts come from a dialect inventory, built by reading the narratives in batches.
AI-drafted from 12 narratives
Under review as I read
“Which woman, born near Gilmer in Upshur County, moved to Marshall the year after the war ended?”
What it testsAsks about something one specific narrative says. Level 2: Same fact, different wording from the text.
A pass looks likeFinds that narrative and answers from it, with a citation.
Found Nancy King at #1.
Came back: Nancy King, Betty Farrow, Eliza Holman, Julia Blanks, Martha Spence Bunton
Answer“Which narrator describes an enslaver who deserted the Confederate army, was hunted down by soldiers, and never came home, except his horse, which returned riderless?”
What it testsAsks about something one specific narrative says. Level 3: The fact is buried mid- or late-narrative.
A pass looks likeFinds that narrative and answers from it, with a citation.
Nancy King was not in the top 5.
Came back: Thomas Cole, Albert Hill, Frank Bell, John Finnely
Answer“Who recounts a relative using a trap strung across a road to unseat mounted Klansmen chasing him?”
What it testsAsks about something one specific narrative says. Level 5: The text describes it without naming it.
A pass looks likeFinds that narrative and answers from it, with a citation.
Nancy King was not in the top 5.
Came back: Lorenza Ezell, Sam Kilgore, Charley Hurt, Agatha Babino
Answer“Why did Nancy King's enslaver volunteer so eagerly to fight for the Confederacy?”
What it testsAsks about something one specific narrative says. Level 7: The question contains a wrong fact.
A pass looks likeCorrects the wrong fact instead of inventing an answer.
Found Nancy King at #1.
Came back: Nancy King, Thomas Cole, Lucinda Elder, John Finnely
Answer“Which woman, enslaved near Jasper, later lived in Beaumont and could only guess she was around 80?”
What it testsAsks about something one specific narrative says. Level 2: Same fact, different wording from the text.
A pass looks likeFinds that narrative and answers from it, with a citation.
Mary Kindred was not in the top 5.
Came back: Ann Hawthorne, Victor Duhon, Mary Johnson
Answer“Whose father was killed in a robbery while carrying the mail, and in what year?”
What it testsAsks about something one specific narrative says. Level 3: The fact is buried mid- or late-narrative.
A pass looks likeFinds that narrative and answers from it, with a citation.
Mary Kindred was not in the top 5.
Came back: Julia Blanks, Sam Kilgore, Unknown, Josie Brown
Answer“Who explains the names people used for sweetening with syrup versus with sugar?”
What it testsAsks about something one specific narrative says. Level 4: The text only says it in dialect.
A pass looks likeFinds that narrative and answers from it, with a citation.
Found Mary Kindred at #2.
Came back: Ellen Betts, Mary Kindred, Ellen Butler, Sarah Ford
Answer“Which narrator describes a grandmother whose healing skills were officially recognized and rewarded with land?”
What it testsAsks about something one specific narrative says. Level 5: The text describes it without naming it.
A pass looks likeFinds that narrative and answers from it, with a citation.
Mary Kindred was not in the top 5.
Came back: Unknown, Betty Farrow, Mother Anne Clark, Harriet Collins
Answer“Which man was still running his own concrete business in Fort Worth in his nineties?”
What it testsAsks about something one specific narrative says. Level 2: Same fact, different wording from the text.
A pass looks likeFinds that narrative and answers from it, with a citation.
Found Sam Kilgore at #1.
Came back: Sam Kilgore, James Cape, James Brown, Lewis Jones, Will Daily
Answer“Who, as a boy in England, went down to the docks to look for cotton bales shipped from home?”
What it testsAsks about something one specific narrative says. Level 3: The fact is buried mid- or late-narrative.
A pass looks likeFinds that narrative and answers from it, with a citation.
Found Sam Kilgore at #1.
Came back: Sam Kilgore, Thomas Cole, William Davis, Zek Brown
Answer“Who said their enslaver kept order by threatening to hand people over to a slave trader instead of whipping them?”
What it testsAsks about something one specific narrative says. Level 4: The text only says it in dialect.
A pass looks likeFinds that narrative and answers from it, with a citation.
Sam Kilgore was not in the top 5.
Came back: Frank Bell, Sarah Ford, Harry Johnson, Louis Fowler, Elige Davison
Answer“Who describes white landowners organizing an armed ambush that ended night-rider attacks in their area?”
What it testsAsks about something one specific narrative says. Level 5: The text describes it without naming it.
A pass looks likeFinds that narrative and answers from it, with a citation.
Found Sam Kilgore at #5.
Came back: Pierce Harper, William Hamilton, Cato Carter, William Branch, Sam Kilgore
Answer“Which wars did Sam Kilgore serve in, and what was his role in each?”
What it testsAsks about something one specific narrative says. Level 6: Needs facts from two parts of the narrative.
A pass looks likeFinds that narrative and answers from it, with a citation.
Found Sam Kilgore at #1.
Came back: Sam Kilgore, William Branch, Zek Brown
Answer“In which major Civil War battle did Sam Kilgore fight?”
What it testsAsks about something one specific narrative says. Level 7: The question contains a wrong fact.
A pass looks likeCorrects the wrong fact instead of inventing an answer.
Found Sam Kilgore at #1.
Came back: Sam Kilgore, John Finnely, Thomas Cole
Answer“Which woman still lived in a cabin on the Brazoria County plantation land where she was born?”
What it testsAsks about something one specific narrative says. Level 2: Same fact, different wording from the text.
A pass looks likeFinds that narrative and answers from it, with a citation.
Pinkie Kelly was not in the top 5.
Came back: Mary Johnson, Martha Spence Bunton, Ann Hawthorne, Molly Harrell
Answer“On which plantation did an enslaved woman run a separate sick building, treating illness with calomel or turpentine?”
What it testsAsks about something one specific narrative says. Level 3: The fact is buried mid- or late-narrative.
A pass looks likeFinds that narrative and answers from it, with a citation.
Pinkie Kelly was not in the top 5.
Came back: Ann Hawthorne, Harriet Barrett, Sarah Benjamin, Mary Kincheon Edwards, Julia Blanks
Answer“Who learned of emancipation only when a stranger on horseback confronted the enslaver as workers were heading to the fields?”
What it testsAsks about something one specific narrative says. Level 3: The fact is buried mid- or late-narrative.
A pass looks likeFinds that narrative and answers from it, with a citation.
Pinkie Kelly was not in the top 5.
Came back: Andy Anderson, Eli Davison, Thomas Cole, Albert Hill
Answer“Who blamed her lifelong bad luck on the phase of the moon when she was born?”
What it testsAsks about something one specific narrative says. Level 4: The text only says it in dialect.
A pass looks likeFinds that narrative and answers from it, with a citation.
Found Pinkie Kelly at #1.
Came back: Aunt Pinkie Kelly, Harriet Collins, Josie Brown, Rosanna Frazier, Mandy Hadnot
Answer“Which man came from a South Carolina tobacco plantation and farmed near Madisonville starting in 1869?”
What it testsAsks about something one specific narrative says. Level 2: Same fact, different wording from the text.
A pass looks likeFinds that narrative and answers from it, with a citation.
Toby Jones was not in the top 5.
Came back: Zek Brown, Monroe Brackins, Anderson And Minerva Edwards, Thomas Cole
Answer“Which narrator describes enslaved people being made to look well-fed before being sold?”
What it testsAsks about something one specific narrative says. Level 3: The fact is buried mid- or late-narrative.
A pass looks likeFinds that narrative and answers from it, with a citation.
Toby Jones was not in the top 5.
Came back: Frank Bell, Adeline Cunningham, Mother Anne Clark, Sarah Ashley, Thomas Cole
Answer“Who escaped continued forced labor after the war by stealing a horse, then sent the horse back home?”
What it testsAsks about something one specific narrative says. Level 5: The text describes it without naming it.
A pass looks likeFinds that narrative and answers from it, with a citation.
Found Toby Jones at #1.
Came back: Toby Jones, Felix Haywood, Thomas Cole, James Cape
Answer“What did Toby Jones say freedom left him with, and how did he and his wife survive their first years in Texas?”
What it testsAsks about something one specific narrative says. Level 6: Needs facts from two parts of the narrative.
A pass looks likeFinds that narrative and answers from it, with a citation.
Found Toby Jones at #1.
Came back: Toby Jones, Thomas Cole, James Cape, Willis Easter, Jack Cauthern
Answer“Why did Toby Jones's enslaver free him right after the war?”
What it testsAsks about something one specific narrative says. Level 7: The question contains a wrong fact.
A pass looks likeCorrects the wrong fact instead of inventing an answer.
Found Toby Jones at #1.
Came back: Toby Jones, Felix Haywood, Frank Bell, Eli Davison
Answer“Which woman, born in Harrison County, kept working for her former enslavers for four years after the war and later lived near Karnack?”
What it testsAsks about something one specific narrative says. Level 2: Same fact, different wording from the text.
A pass looks likeFinds that narrative and answers from it, with a citation.
Found Lizzie Jones at #1.
Came back: Lizzie Jones, Sarah Allen, Mary Armstrong, Mary Johnson, Nancy King
Answer“On which plantation could couples marry only after a certain age, and never with someone from another plantation?”
What it testsAsks about something one specific narrative says. Level 3: The fact is buried mid- or late-narrative.
A pass looks likeFinds that narrative and answers from it, with a citation.
Found Lizzie Jones at #5.
Came back: Ellen Betts, Lucinda Elder, Jeff Calhoun, Phoebe Henderson, Lizzie Jones
Answer“Which narrator was given a nickname by her enslavers because of how dark and shiny her skin was?”
What it testsAsks about something one specific narrative says. Level 4: The text only says it in dialect.
A pass looks likeFinds that narrative and answers from it, with a citation.
Lizzie Jones was not in the top 5.
Came back: Mother Anne Clark, Jacob Branch, Josephine Howard, Harriet Collins, Mattie Gilmore
Answer“Who describes an enslaver deliberately hiding emancipation and telling his family to keep people working as long as possible?”
What it testsAsks about something one specific narrative says. Level 5: The text describes it without naming it.
A pass looks likeFinds that narrative and answers from it, with a citation.
Lizzie Jones was not in the top 5.
Came back: Molly Harrell, Frank Bell, Sarah Ford, Elige Davison, Eli Coleman
Answer“How old was Lizzie Jones when interviewed, and when was she born?”
What it testsAsks about something one specific narrative says. Level 6: Needs facts from two parts of the narrative.
A pass looks likeReports both conflicting figures instead of picking one.
Found Lizzie Jones at #1.
Came back: Lizzie Jones, Liza Jones, Lizzie Hughes, Lewis Jones
Answer“Which woman in Beaumont cared for a husband of sixty years who could no longer see, hear, or speak?”
What it testsAsks about something one specific narrative says. Level 2: Same fact, different wording from the text.
A pass looks likeFinds that narrative and answers from it, with a citation.
Liza Jones was not in the top 5.
Came back: Mary Johnson, Victor Duhon, Chris Franklin, Joe Barnes, Unknown
Answer“Which narrator told interviewers she had already seen them in a dream the night before they came?”
What it testsAsks about something one specific narrative says. Level 3: The fact is buried mid- or late-narrative.
A pass looks likeFinds that narrative and answers from it, with a citation.
Liza Jones was not in the top 5.
Came back: Mother Anne Clark, Julia Blanks, Rosa Green, John Day, John Barker
Answer“Who found large amounts of cash hidden in household containers while keeping house for a white employer?”
What it testsAsks about something one specific narrative says. Level 3: The fact is buried mid- or late-narrative.
A pass looks likeFinds that narrative and answers from it, with a citation.
Found Liza Jones at #1.
Came back: Liza Jones, Mother Anne Clark, Julia Blanks, Unknown
Answer“What special titles did Liza Jones say her mother and father held on the plantation?”
What it testsAsks about something one specific narrative says. Level 4: The text only says it in dialect.
A pass looks likeFinds that narrative and answers from it, with a citation.
Found Liza Jones at #1.
Came back: Liza Jones, William Davis, Silvia King, Virginia Bell, Lewis Jones
Answer“Who describes a parent disarming an enslaver and locking him up to keep him out of trouble when Union soldiers arrived?”
What it testsAsks about something one specific narrative says. Level 5: The text describes it without naming it.
A pass looks likeFinds that narrative and answers from it, with a citation.
Liza Jones was not in the top 5.
Came back: John Day, Molly Harrell, Frank Bell, Zek Brown
Answer“Which man worked at a cotton gin in La Grange and later for Armour in Fort Worth?”
What it testsAsks about something one specific narrative says. Level 2: Same fact, different wording from the text.
A pass looks likeFinds that narrative and answers from it, with a citation.
Found Lewis Jones at #1.
Came back: Lewis Jones, Sam Kilgore, Zek Brown, John Finnely, James Cape
Answer“How was freedom announced on the Tate plantation?”
What it testsAsks about something one specific narrative says. Level 2: Same fact, different wording from the text.
A pass looks likeFinds that narrative and answers from it, with a citation.
Found Lewis Jones at #3.
Came back: John James, Lewis Jones, Laura Cornish, Sarah Ford
Answer“Who kept control of a runaway carriage team frightened by a newly installed steam whistle?”
What it testsAsks about something one specific narrative says. Level 3: The fact is buried mid- or late-narrative.
A pass looks likeFinds that narrative and answers from it, with a citation.
Found Lewis Jones at #2.
Came back: Carey Davenport, Lewis Jones, Martin Jackson, Ben Kinchlow, James Cape
Answer“On which plantation did dancers compete while balancing a glass of water on their heads?”
What it testsAsks about something one specific narrative says. Level 3: The fact is buried mid- or late-narrative.
A pass looks likeFinds that narrative and answers from it, with a citation.
Lewis Jones was not in the top 5.
Came back: Preely Coleman, Ellen Betts, William M Adams, John Finnely, Harriet Collins
Answer“Which narrator says his father was used by the enslaver to father children with many women?”
What it testsAsks about something one specific narrative says. Level 5: The text describes it without naming it.
A pass looks likeFinds that narrative and answers from it, with a citation.
Lewis Jones was not in the top 5.
Came back: Frank Bell, Sarah Ford, Thomas Cole, Thomas Johns, Laura Cornish
Answer“How long did Lewis Jones work at the La Grange cotton gin?”
What it testsAsks about something one specific narrative says. Level 6: Needs facts from two parts of the narrative.
A pass looks likeReports both conflicting figures instead of picking one.
Found Lewis Jones at #1.
Came back: Lewis Jones, Sam Kilgore, Liza Jones
Answer“Whose parents were brought from North Carolina to Red River County, near Clarksville?”
What it testsAsks about something one specific narrative says. Level 2: Same fact, different wording from the text.
A pass looks likeFinds that narrative and answers from it, with a citation.
Found Harriet Jones at #1.
Came back: Harriet Jones, Nelsen Denson, Julia Blanks, Ben Kinchlow, William Davis
Answer“Who describes two strangers who stayed for weeks pretending to look for land and turned out to be Union spies?”
What it testsAsks about something one specific narrative says. Level 3: The fact is buried mid- or late-narrative.
A pass looks likeFinds that narrative and answers from it, with a citation.
Harriet Jones was not in the top 5.
Came back: Thomas Cole, Cato Carter, John Finnely, John Day
Answer“Through which town was cotton from the Red River County plantation shipped to market, and how far away was it?”
What it testsAsks about something one specific narrative says. Level 4: The text only says it in dialect.
A pass looks likeFinds that narrative and answers from it, with a citation.
Harriet Jones was not in the top 5.
Came back: Ben Kinchlow, Nelsen Denson, Harrison Boyd
Answer“Describe Harriet Jones's own wedding: when, where, what she wore, and what her mother gave her.”
What it testsAsks about something one specific narrative says. Level 6: Needs facts from two parts of the narrative.
A pass looks likeFinds that narrative and answers from it, with a citation.
Found Harriet Jones at #1.
Came back: Harriet Jones, Julia Blanks, Lizzie Jones, Mandy Hadnot
Answer“Which man, born free among the Choctaw, settled in Waco in 1874?”
What it testsAsks about something one specific narrative says. Level 2: Same fact, different wording from the text.
A pass looks likeFinds that narrative and answers from it, with a citation.
Found Spence Johnson at #3.
Came back: Chris Franklin, Ben Kinchlow, Spence Johnson, Zek Brown, Ann J Edwards
Answer“What price did Spence Johnson's mother and her children bring at auction?”
What it testsAsks about something one specific narrative says. Level 3: The fact is buried mid- or late-narrative.
A pass looks likeFinds that narrative and answers from it, with a citation.
Found Spence Johnson at #1.
Came back: Spence Johnson, James D Johnson, Martha Spence Bunton, Pauline Johnson
Answer“Which narrator describes being kidnapped from free territory as a toddler and sold into slavery?”
What it testsAsks about something one specific narrative says. Level 5: The text describes it without naming it.
A pass looks likeFinds that narrative and answers from it, with a citation.
Spence Johnson was not in the top 5.
Came back: Frank Bell, Mother Anne Clark, Thomas Cole, Francis Black, Josephine Howard
Answer“Who tells of a man repeatedly sold away and stolen back by the same white man until he was found dead?”
What it testsAsks about something one specific narrative says. Level 5: The text describes it without naming it.
A pass looks likeFinds that narrative and answers from it, with a citation.
Spence Johnson was not in the top 5.
Came back: Frank Bell, John Finnely, Jerry Boykins, Preely Coleman
Answer“Why did Spence Johnson's family decide to move from the Choctaw Nation to Louisiana?”
What it testsAsks about something one specific narrative says. Level 7: The question contains a wrong fact.
A pass looks likeCorrects the wrong fact instead of inventing an answer.
Found Spence Johnson at #1.
Came back: Spence Johnson, Pauline Johnson, Gus Johnson, James D Johnson
Answer“Which woman said she was born in Morocco and later lived in Marlin?”
What it testsAsks about something one specific narrative says. Level 2: Same fact, different wording from the text.
A pass looks likeFinds that narrative and answers from it, with a citation.
Found Silvia King at #1.
Came back: Silvia King, Olivier Blanchard, Agatha Babino
Answer“Who described a ritual using ashes from a lightning-struck pine tree to make legal trouble go away?”
What it testsAsks about something one specific narrative says. Level 3: The fact is buried mid- or late-narrative.
A pass looks likeFinds that narrative and answers from it, with a citation.
Found Silvia King at #1.
Came back: Silvia King, Willis Easter, Harriet Collins, Rosanna Frazier
Answer“Which narrator said she grew new teeth late in life?”
What it testsAsks about something one specific narrative says. Level 3: The fact is buried mid- or late-narrative.
A pass looks likeFinds that narrative and answers from it, with a citation.
Silvia King was not in the top 5.
Came back: Mary Johnson, Betty Farrow, Mother Anne Clark, Josephine Howard, Unknown
Answer“Who describes being taken from her husband and children overseas, drugged, and shipped to America?”
What it testsAsks about something one specific narrative says. Level 5: The text describes it without naming it.
A pass looks likeFinds that narrative and answers from it, with a citation.
Found Silvia King at #1.
Came back: Silvia King, Mother Anne Clark, Francis Black, Mary Armstrong, Julia Blanks
Answer“Which woman was paired with a man against her will and fought him until threatened with a whip?”
What it testsAsks about something one specific narrative says. Level 5: The text describes it without naming it.
A pass looks likeFinds that narrative and answers from it, with a citation.
Found Silvia King at #2.
Came back: John Finnely, Silvia King, Mother Anne Clark
Answer“How did Silvia King reunite with her husband and children after emancipation?”
What it testsAsks about something one specific narrative says. Level 7: The question contains a wrong fact.
A pass looks likeCorrects the wrong fact instead of inventing an answer.
Found Silvia King at #1.
Came back: Silvia King, Julia Blanks, Phoebe Henderson, Sarah Allen
Answer“Which man, born in Wharton County, spent his childhood in Matamoros after his mother was freed?”
What it testsAsks about something one specific narrative says. Level 2: Same fact, different wording from the text.
A pass looks likeFinds that narrative and answers from it, with a citation.
Found Ben Kinchlow at #1.
Came back: Ben Kinchlow, Monroe Brackins, Martin Jackson
Answer“Who worked as an unpaid scout for Texas Ranger Captain McNelly and supported himself by gambling?”
What it testsAsks about something one specific narrative says. Level 3: The fact is buried mid- or late-narrative.
A pass looks likeFinds that narrative and answers from it, with a citation.
Found Ben Kinchlow at #1.
Came back: Ben Kinchlow, James Cape
Answer“Who describes enslaved teamsters hauling cotton to the border and choosing to cross into Mexico instead of returning?”
What it testsAsks about something one specific narrative says. Level 5: The text describes it without naming it.
A pass looks likeFinds that narrative and answers from it, with a citation.
Found Ben Kinchlow at #3.
Came back: Felix Haywood, Jacob Branch, Ben Kinchlow, James Boyd
Answer“Who described seeing electricity playing across cattle horns and horses' manes during storms?”
What it testsAsks about something one specific narrative says. Level 5: The text describes it without naming it.
A pass looks likeFinds that narrative and answers from it, with a citation.
Found Ben Kinchlow at #4.
Came back: Monroe Brackins, Tempie Cummins, John Barker, Ben Kinchlow, Harrison Beckett
Answer“What happened in the fight on Palo Alto Prairie, and what was Ben Kinchlow's role?”
What it testsAsks about something one specific narrative says. Level 6: Needs facts from two parts of the narrative.
A pass looks likeReports both conflicting figures instead of picking one.
Found Ben Kinchlow at #1.
Came back: Ben Kinchlow, John Finnely, Louis Fowler
Answer“Which radio story about Captain McNelly did Ben Kinchlow confirm?”
What it testsAsks about something one specific narrative says. Level 7: The question contains a wrong fact.
A pass looks likeCorrects the wrong fact instead of inventing an answer.
Found Ben Kinchlow at #1.
Came back: Ben Kinchlow, William Branch
Answer“How did people in these narratives learn they were free, and how quickly?”
What it testsAsks about a theme that runs across several narratives.
A pass looks likeBrings back relevant accounts from several narrators and cites each one.
Of the narratives this question was drafted from, 0 of 6 came back. Others may be just as relevant.
Came back: Julia Blanks, Sarah Ford, Molly Harrell, Tempie Cummins
Answer“What did people drink when real coffee wasn't available?”
What it testsAsks about a theme that runs across several narratives.
A pass looks likeBrings back relevant accounts from several narrators and cites each one.
Of the narratives this question was drafted from, 2 of 2 came back. Others may be just as relevant.
Came back: Valmar Cormier, Mary Kindred, Sarah Ford, Ben Kinchlow
Answer“Accounts of confronting or outwitting the Klan or night riders”
What it testsAsks about a theme that runs across several narratives.
A pass looks likeBrings back relevant accounts from several narrators and cites each one.
Of the narratives this question was drafted from, 1 of 2 came back. Others may be just as relevant.
Came back: Lorenza Ezell, Sam Kilgore, Pierce Harper
Answer“How was illness treated on plantations?”
What it testsAsks about a theme that runs across several narratives.
A pass looks likeBrings back relevant accounts from several narrators and cites each one.
Of the narratives this question was drafted from, 0 of 4 came back. Others may be just as relevant.
Came back: Ann Hawthorne, Mary Armstrong, Harriet Collins, Elige Davison, Harriet Barrett
Answer“Wedding customs during and after slavery”
What it testsAsks about a theme that runs across several narratives.
A pass looks likeBrings back relevant accounts from several narrators and cites each one.
Of the narratives this question was drafted from, 1 of 5 came back. Others may be just as relevant.
Came back: Ann Hawthorne, Jeff Calhoun, Richard Carruthers, Harriet Jones
Answer“People born outside the United States or outside slavery”
What it testsAsks about a theme that runs across several narratives.
A pass looks likeBrings back relevant accounts from several narrators and cites each one.
Of the narratives this question was drafted from, 0 of 4 came back. Others may be just as relevant.
Came back: Gus Johnson, Jeptha Choice, Felix Haywood, Henry H Buttler, Julia Blanks
Answer“Folk beliefs, charms, and conjure”
What it testsAsks about a theme that runs across several narratives.
A pass looks likeBrings back relevant accounts from several narrators and cites each one.
Of the narratives this question was drafted from, 0 of 6 came back. Others may be just as relevant.
Came back: William Adams, Willis Easter, Rosanna Frazier, Harriet Collins
Answer“What happened when patrols caught enslaved people off the plantation without a pass?”
What it testsAsks the same thing twice: once in plain English, once in the transcript's phonetic spelling.
A pass looks likeBoth phrasings reach the same passages.
Toby Jones was not in the top 5.
Came back: Green Cumby, Harry Johnson, John James, Fred Brown, Pierce Harper
Answer“What did the patterrollers do when they cotched folks without a pass?”
What it testsAsks the same thing twice: once in plain English, once in the transcript's phonetic spelling.
A pass looks likeBoth phrasings reach the same passages.
Toby Jones was not in the top 5.
Came back: Green Cumby, Harry Johnson, Ben Kinchlow, Fred Brown
Answer“Horses frightened and running away with a carriage”
What it testsAsks the same thing twice: once in plain English, once in the transcript's phonetic spelling.
A pass looks likeBoth phrasings reach the same passages.
Lewis Jones was not in the top 5.
Came back: Albert Hill, Carey Davenport, Julia Francis Daniels, Cato Carter, Green Cumby
Answer“De hosses skeert and run off with de coach”
What it testsAsks the same thing twice: once in plain English, once in the transcript's phonetic spelling.
A pass looks likeBoth phrasings reach the same passages.
Found Lewis Jones at #1.
Came back: Lewis Jones, Andy Anderson, James Brown, William Byrd
Answer“Were enslaved people in Texas treated well?”
What it testsInvites a sweeping generalization about slavery.
A pass looks likeReports individual accounts instead of one verdict.
Of the narratives this question was drafted from, 0 of 6 came back. Others may be just as relevant.
Came back: Jack Bess, Adeline Cunningham, Jack Cauthern, Harry Johnson, Andy Anderson
Answer“Did formerly enslaved people prefer slavery to freedom?”
What it testsInvites a sweeping generalization about slavery.
A pass looks likeReports individual accounts instead of one verdict.
Of the narratives this question was drafted from, 1 of 2 came back. Others may be just as relevant.
Came back: Felix Haywood, Thomas Johns, Armstead Barrett, Toby Jones, Eli Davison
Answer“What does Harriet Jones's account tell us about slavery as a whole?”
What it testsInvites a sweeping generalization about slavery.
A pass looks likeReports individual accounts instead of one verdict.
Of the narratives this question was drafted from, 1 of 1 came back. Others may be just as relevant.
Came back: Silvia King, Harriet Jones, Frank Bell, Elige Davison
Answer“What did the narrators say about the attack on Pearl Harbor?”
What it testsOut of scope, or needs counting across the whole collection.
A pass looks likeSays the passages can't answer it, instead of answering from general knowledge.
The answer declined.
Came back: Jacob Branch, Louis Fowler, Felix Haywood, John Day
Answer“What was Lincoln's Gettysburg Address about?”
What it testsOut of scope, or needs counting across the whole collection.
A pass looks likeSays the passages can't answer it, instead of answering from general knowledge.
The answer declined.
Came back: John Finnely, Felix Haywood, Ann J Edwards, William M Adams
Answer“What did Sam Kilgore say about the 1918 flu pandemic?”
What it testsOut of scope, or needs counting across the whole collection.
A pass looks likeSays the passages can't answer it, instead of answering from general knowledge.
The answer declined.
Came back: Sam Kilgore, Mary Armstrong, Thomas Cole, Eli Davison
Answer“How many narratives in the collection mention the Ku Klux Klan?”
What it testsOut of scope, or needs counting across the whole collection.
A pass looks likeSays the passages can't answer it, instead of answering from general knowledge.
The answer declined.
Came back: Lorenza Ezell, Agatha Babino, William Hamilton, Pierce Harper, John James
AnswerI ran all 78 through the same pipeline the demo uses. Nothing here is final, but a few patterns were hard to miss:
I'm working through the narratives in batches, noting how dialect actually shows up: respelled everyday words, period vocabulary, and grammar. That review will reshape the test questions, especially the dialect pairs, so that a plain question and its phonetic twin ask exactly the same thing. This is a side project I'm learning through, so it'll move in steps. I'll update this page as it does.
The design question here was trust, not fluency. A fluent answer about this material is easy to produce and easy to get wrong, so the interface shows what was found, how closely it matched, and who said it before any generated text appears. If the evidence and the answer disagree, a visitor can see it.

I started as a software engineer at Dell, building a UI component library used across hundreds of IT Services projects. At Disney, I moved into UX engineering, building prototypes, a component library, and a metrics dashboard for the CTO. At H-E-B, I started as a Design Technologist II and now work as a senior design technologist on a lean, agile team that explores emerging technologies and turns them into tools, prototypes, and new ways of working.
That background shapes how I work: I make practical trade-offs between fidelity, speed, and learning, and I care about turning ambiguous ideas into something people can test and react to.
Resume ↗