Applied AI · customer-facing builds · Austin, remote

I build AI into the workflows customers already run, and I publish how often it breaks.

Eight years running a $14M+ enterprise book at GLG, where I was the AI person on a non-AI team, and four builds with dated evals on every one, so the log on the right is the last run.

See the builds ↓
00:00.0

In plain terms:

Agents · Field discovery · LLM pipeline, built solo

Sixty seconds in the car becomes a record onboarding can open.

A rep talks into their phone in a parking lot, and onboarding gets a record with the unanswered questions named on it.

In plain terms: a salesperson rambles into their phone after a visit, and the tool pulls out what matters, flags what nobody asked about, and waits for a person to okay it before anything touches the sales system.

What the rep actually said, four hours later

The sentence that decides the go-live date is the last one, after the tacos.
1 · Talk
Kestrel · capture0:00
Sixty seconds, one hand, in the car.
2 · Match
A fixed list of questions, and the AI cannot invent one.
3 · Confirm
Rental counter on a second systemconfirm before onboarding
Decision maker · Ted?
Never came up: budget, timeline, chemical licence
4 · Hand off
Plain CRM fields, a quote under every answer.

Ten planted instructions, five runs each, forty of fifty unchanged.

unchangedmoved a proposal
Totals from the 22 Sep run · which runs moved is laid out for the mock

In plain terms: I hid sneaky instructions in the notes to see if I could trick it, and it held steady most of the time but slipped often enough that a person still approves every change.

RAG + MCP · Life in Pixels · six months of my own daily data

Ask it a question in plain language, and every day it cites is checked before you see the answer.

In plain terms: it works like a search engine for my own daily log that has to show which days the answer came from, and it double-checks those days are real before it answers.

a replayed run
?

shown
MCP · the same pipeline, as a local server
Claude Desktoppixels-rag · stdiodaily data · on disk
Two read-only tools, the same router and validator, and the data never leaves the machine.

In plain terms: this lets Claude on my laptop ask the same questions directly, and my data never gets uploaded anywhere.

0/26valid citations, 26 questions
3routes: search, filter, sum
Replay all 26 questions →
Fine-tuning · in progress

Card matching, tuned against prompted, on the same golden set.

Every model call on this page is a prompted Haiku or Sonnet under a schema, and the next build is a small tuned model on the one task that is a closed classification, which may end with keep prompting, and that goes here either way.

In plain terms: I am testing whether training a small model on one narrow job beats just asking a big model well, and if it does not, I will say so here.

prompted Haiku
prompted Sonnet
tuned small model
bars are placeholders until the scores publish together
Structured outputs · Signal · LLM app, in useTry it on your own notes ↗

Messy documents go in and a read you can act on comes out.

Haiku summarizes each document and Sonnet synthesizes, which cut the main call from roughly 23k to about 1k input tokens, and every section of the brief names its sources.

In plain terms: you drop in a pile of messy account notes and get back a short read on whether the customer is at risk, who actually matters, and the one thing to do today.

What you paste in
What comes back, 51 seconds later

A mood with no event behind it is read right in seven of twenty.

ablation, 20 runs per arm · which runs is laid out for the mock

In plain terms: it is good at spotting trouble tied to something that happened and still weak at reading a vague mood with nothing behind it, which is why that number is on the page.

~23kinput tokens, from ~23k
60minutes of prep
0spaste to read
Structured outputs · Brain DumpTry it here ↗

Five minutes in, four piles out.

The system prompt is assembled from the energy state you pick, and the model sorts every item into one of four buckets under a strict JSON contract, so change the state and watch the same dump re-sort.

In plain terms: you dump everything on your mind, say how you are feeling, and it sorts the pile into what to do today, what can wait, what to keep, and what to let go.

Raw dump · five minutes, no editing

highlighted as the model reads it, one bucket per item, strict JSON
How are you right now?

the five states and their caps are the product's own · the piles are predicted from its rules until the five-state run is recorded

The analysis work

both start from a question I actually had
tests passing 0/35data flowing left to right

In plain terms: raw grocery orders go in on the left, get cleaned and joined in steps, and come out on the right as tidy tables, with 35 automatic checks making sure nothing broke along the way.

Instacart · 3.4M orders

I rebuilt one cited number until it split in half.

0.60 pooled
new · 0.221
veteran · 0.670

New shoppers reorder at 0.221 and veterans at 0.670, so the 0.60 everyone cites describes neither, and a random forest at 0.9886 AUC for veterans against 0.8566 for new users confirmed the split was real.

In plain terms: people who are new to the app rarely buy the same thing twice and regulars almost always do, so the one average everyone quotes is really two very different groups mashed together and it fits neither of them.

See the models ↗
ATX Foodie · Socrata API · 21,160 records · 84 brands

Venues score about two points higher by their fifteenth inspection than at their first.

90.5 · 1st visit92.6 · 15th↑ fewer↓ more
inspection 1up is fewer violations15

In plain terms: restaurants tend to score a little better the more times they get inspected, though the places still open for a fifteenth visit may just be the ones that were already doing fine.

See the findings ↗

Experience

GLGAustin, TX · Remote · Jul 2018 – present 8 years · 6 roles · IC to people manager

In plain terms: I spent eight years running big customer accounts and then the teams that run them, which is where I learned what customers actually need before I started building for them.

2018202020222024now

Skills

five areas

Certs

each one verifiable
Off the clock

The same habit, pointed at things that do not pay.

Contact

Come say howdy.

Open to anything where a messy workflow needs a system built around it.

© 2026 Samie Vargas Austin, TX · Remote · no framework, no build stepSite toolkit ↗