AI engineering
Getting a model to do useful work you can check.
Typed tools on the network's own records, with evidence on every answer.
- Ask, in your own words
- The network's own credits and connections
- Matches, each with its evidence
- Introduction, only after your yes
- Getting a model to do useful work you can check.
- Getting the right data in, clean and connected.
- Fitting outside services together so the whole job gets done.
- 17typed tools
- 5rounds at most
Engineering detail
Stack
- FastAPI
- Pydantic
- PostgreSQL
- pg_trgm
- Supabase
- Next.js 16
- React 19
- pytest, fake model
How a question runs
- Checks first: sign-in, an on and off switch and rate limits, before any model is called.
- Rules first: a lookup runs before the model, so it starts from real data. Names are matched fuzzily; search walks up to four connections.
- The tool loop: seventeen typed tools, up to five rounds, then it must answer. The server checks what you may see on every call.
- Up to five cards, each showing the credits behind that match.
- It proposes an introduction. Nothing is sent until you say yes.
What's built
- A rule-based router before the model
- Seventeen typed tools, five rounds
- Per-message choice of agent tools
- Introductions only after the member's yes
- Card sentences checked by code
- Evaluation harness, being brought online
Data engineering
Getting the right data in, clean and connected.
Credits come from many sources and each must land once, on the right entity.
- Intent
- Momentum
- Relationships
- Entities
- 4ways to place records
- 3truth tiers
Engineering detail
Stack
- PostgreSQL 16
- Supabase
- pg_trgm
- Python
- FastAPI
- asyncpg
- APScheduler jobs, leases
- Hypothesis tests
How a record finds its place
- Version 1 records keep their ids, so each lands once; running the move again changes nothing; the final switch has not happened yet.
- A near match on the name becomes a proposal for a person to review.
What's built
- A graph of people, projects, venues and labels
- Truth tiers: confirmed, observed, inferred
- Matching rules per record type
- Review queue for near duplicates
- Imports from catalogs and public pages
- The move from version 1, with dry runs
Solutions engineering
Fitting outside services together so the whole job gets done.
Cal.com runs the calendar, Stripe the money, and the creator picks the video app.
Free call. Cal.com books the creator's time; the call runs in their video app.Paid call. Stripe holds the card, captures on confirmation, and pays the creator after.Canceled. The booker is refunded by rule, through the same ledger.
Tested against mocks; no paid booking has completed yet.
- 3money steps
- 4video apps
- 8sweep jobs
Engineering detail
Stack
- Stripe
- PaymentIntents, manual capture
- Connect Express
- webhooks
- Cal.com
- Resend
What's built
- Scheduling through each creator's Cal.com
- Escrow: authorize, capture, release
- Refunds, reversals and sweep jobs
- Creator payouts through Stripe Connect
- A video app the creator picks
What this means for your system
A Production Triage ranks the risks in your system in three business days.
See all pricesStart with fifteen minutes.
A free call. You leave knowing which engagement fits, or that none does.
