VOLUME
47
Engagements shipped to production
FRONTIER FIRM · EST. 2024We build AI products
We build AI products
on solid ground.
A small lab of senior engineers shipping production AI for SaaS companies. Eight weeks from problem to live, with no framework lock-in, no orchestration vendors, and no slide decks.
KL · 047 ENGAGEMENTSNEW YORK · BROOKLYN002 / Q3 ′25 OPEN
●TYPESCRIPT●POSTGRES●PGVECTOR●LLM●RAG●EVAL●OPENTELEMETRY●LANGSMITH●ANTHROPIC●OPENAI●NEXT●REDIS●DRIZZLE●ZOD●TRPC●PYTHON●CLICKHOUSE●DATADOG●SENTRY●CLOUDFLARE●VLLM●WEBSOCKET●GRPC●TANSTACK
LICENSE COST
$0
Framework licenses, ever
AVAILABILITY
2/ Q3
Slots open right now
NuforceSeries-B insurer
Affidavit MappCapital markets
AICoachVertical SaaS
ConfidentialB2B marketplace
ConfidentialEHR add-on
PRACTICE · 03 / 03
Three things hold every build up.
Pick the practice closest to your blocker. The three carry every engagement we ship.
PRACTICE · A01 / 03
Direct to the model.
We integrate against the LLM API the same way we integrate against Postgres. No agent frameworks, no orchestration vendors, no migrations every six months.
→ Postgres · queues · edge · the model API
PRACTICE · B02 / 03
Senior, all the way down.
Every line of code on your project is written by an engineer who has shipped before. There is no junior team behind the curtain, no staffing pyramid.
→ 12 senior engineers · 0 juniors · 0 pyramid
PRACTICE · C03 / 03
You own the code.
The repo, the prompts, the evals, the runbook — your team owns everything at handoff. No black boxes, no vendor lock-in, no business held hostage.
→ Full repo · full evals · full runbook · MIT
▲ FROM THE FIELD · CLAIMS-TRIAGE
A live system, week seven of eight.
One of three engagements shipping this quarter. Direct LLM integration over a Series-B insurer's legacy claims pipeline.
Direct to the model.
Boring infrastructure.
Senior people.
We integrate against the LLM API the way you integrate against Postgres. No agent frameworks, no orchestration vendors, no migrations every six months.
−80%
cost on human tasks
+60%
throughput, day 90
0
frameworks shipped
● LIVE · claims-triage.kensink.runBUILD WK 07 / 08
73%
●Auto-resolved
412ms
Median latency
98.2%
●Eval pass rate
Build progress87.5%
ENGAGEMENT SHAPES · 03 / 03
What we ship in eight weeks.
Three engagement shapes. Fixed scope, fixed fee, and real software. You own everything we leave behind.
01 · OPERATIONS
Triage & deflection systems
Tickets, claims, support queues. We sort, summarize, and auto-resolve the long tail; humans keep the hard cases.
- StackPostgres · Python
- Build6–8 weeks
- Typical lift40–75% deflection
Fixed scope · fixed fee
02 · INTERNAL
Research & analyst copilots
Retrieval over your knowledge base, threaded into the tools your team already opens at 9 a.m. on Monday.
- Stackpgvector · Next
- Build8–10 weeks
- Typical lift3–5 hrs / analyst / wk
Fixed scope · fixed fee
03 · PRODUCT
Customer-facing AI features
A model-powered slice inside your product, with the eval suite, observability, and runbook your team can extend.
- StackTS · pgvector
- Build8–12 weeks
- Typical lift+18% activation
Fixed scope · fixed fee
Bring the shape that fits your build. We’ll come back with a written scope, fixed fee, and a start week.
Pick the shape that fits →Numbers from forty-seven shipped engagements.
MEASURED · WEIGHTED · 2024–2026TIME TO PROD
8wk
Median time from kickoff to a build serving production traffic.
COST REDUCTION
−80%
Average reduction in human-task cost across our last twelve engagements.
PRODUCTIVITY
+60%
Mean productivity lift measured at the team level, ninety days post-launch.
FIRST-YEAR ROI
340%
Average first-year ROI, weighted by engagement size, on closed projects.
A four-stage build, no theatre.
Each stage has one deliverable and one decision point. No status meetings and no slide decks. Just the next thing, on a Tuesday.
WK 01 · DISCOVERY
Scope memo
Two senior architects on-site for a week. The output is a written brief naming the smallest valuable thing to ship.
on-sitememo
WK 02–06 · BUILD
Production system
Code, prompts, evals, infrastructure. Direct LLM integration on your stack, on your data, behind your auth.
postgrespythonts
WK 07 · EVAL
Behavior tests
A test suite for behavior, not just unit tests. Dashboards your team will actually open on Monday morning.
braintrustgrafana
WK 08 · HANDOFF
Runbook + pairing
A week of pairing with your team. We leave a system your engineers can extend, not a binary they can't.
pairing90-day warranty
◆ SELECTED WORK · 12 OF 47
Recent production systems.
Five of the last twelve engagements. Each one is a real system running today, with measured outcomes.
No. 01Claims triage engineSERIES-B INSURER73% auto-resolveApr 2026Read →No. 02Internal-research copilotCAPITAL MARKETS · 200 ANALYSTS4 hrs/analyst savedMar 2026Read →No. 03Pricing-recommendation systemB2B MARKETPLACE+4.2 pt marginFeb 2026Read →No. 04Support-deflection layerVERTICAL SAAS41% deflectedJan 2026Read →No. 05Clinical-summary toolEHR ADD-ON VENDORFDA-cleared pathNov 2025Read →
Yes, forty-seven times so far. The trick is that we scope the smallest valuable thing in week one, then defer everything else to the runbook. We will turn down work that we can't scope into eight weeks.
Then we shape it as a multi-phase program (quarterly milestones, fixed price per phase) or an ongoing partnership (fractional senior engineering for post-launch teams). Same evaluation discipline, same source ownership at every milestone. Sprints are the lead promise; they're not the only one.
Less surface area, fewer migrations, lower bills. We integrate against the LLM API the same way we integrate against Postgres. It's a tool, not a platform. Our clients keep that bet for years.
You own the code, the prompts, the evals, and the runbook on day one. There is no license, no lock-in, and no retainer that bills past cutover.
Two senior engineers, named, on the contract. There is no junior team behind the curtain. The people you meet on day one are the people writing every line.
We carry a 90-day warranty. If a regression ships from our code, we fix it on our dime, with no retainer and no rebill.
Bring a real problem.
We'll bring code on day one.
Book a 15-min intro →
Q3 ′25 · 2 SLOTS
FIXED SCOPE · 8 WEEKS
LEAD TIME 3–4 WEEKS