Services
Work with the lab
Small, fixed-scope projects. Each one ends with software you own, the docs to run it, and a system that keeps working when nobody is watching it.
Agent workflows with a sign-off
An AI agent takes over a recurring job, like reading an inbox, drafting updates or filing records, and a person approves each change with one tap. There’s no new database or admin app to keep alive.
A good fit if
- Someone does the same task by hand every day
- Changes have to be reviewed before they go out
- You’ve been quoted a custom internal tool
You get
- The agent and its approval step
- Hard limits on what it can do
- A record of every decision and the reason for it
- A runbook your team can follow
- Inbox or queuewhere the work arrives
- Agent draftsthe exact change
- Guardrailslimits it can’t argue with
- Your sign-offone tap
- Appliedand logged
Data feeds and change monitoring
A source you care about, whether public filings, a vendor export or your own system, gets snapshotted on a schedule and compared run over run. What changed is ranked and delivered, and the history is kept even when the source doesn’t keep it.
A good fit if
- Someone checks a site or spreadsheet for changes by hand
- The source overwrites itself and loses history
- You need to hear about a change within a day, not a quarter
You get
- The pipeline, running on a schedule
- An archive of every snapshot
- A ranked feed, digest or alerts
- A warning when a source changes shape
- Sourceany schedule
- Snapshotkept forever
- Sanity checkbad pulls are skipped
- Diff + rankwhat actually matters
- Feeddigest or alerts
Model evaluation
Before an AI model does real work in your domain, it gets tested on questions it hasn’t seen, with the answers fixed ahead of time, alongside the other models you’re considering.
A good fit if
- You’re choosing between model vendors
- A model upgrade needs checking before it ships
- Someone has to sign off that the AI is good enough
You get
- A question set you own
- A harness that reruns it on any model
- A scored comparison
- A plain-language read of where each model fails
- Questionsfrom your domain
- Freezehashed before any run
- Runevery model the same way
- Scoreharness + judge
- Hand checkevery rubric row
- Reportwhat to trust
Data migrations
Records get moved from one system to another with mapping, cleanup, test loads and reconciliation, so every record is accounted for before cutover. If the old and new systems have to keep talking afterward, that integration is part of the job. This is the oldest skill in the shop.
A good fit if
- You’re replacing a CRM or core system
- Two companies’ data has to become one
- A past migration left records nobody trusts
- Two systems have to stay in sync after go-live
You get
- Repeatable migration scripts
- Reconciliation reports, record by record
- A cutover plan with a way back
- Clean data on the other side
- Old systemas it really is
- Map + cleanfield by field
- Test loadsas many as it takes
- Reconcileevery record counted
- Cutoverwith a rollback
How it runs
Every engagement- 01
A short call
What the problem is, what it costs you today, and whether it’s a fit. If it isn’t, you’ll hear that on the call.
- 02
Scoping
A written scope and a fixed price for each phase. You know the total before anything gets built.
- 03
Build in phases
Working software at the end of every phase, running on your data. You can stop after any phase and keep what’s done.
- 04
Handoff
The code in your repository, the docs, a runbook and a walkthrough. Nothing to license, and nothing that stops running if we part ways.
Start with a note