Phase 02 — Build

Ship production AI in weeks.

Senior engineers, fixed scope, weekly demos. We build order automation, supplier integrations, quoting assistants, and internal AI tools to production grade, with security, monitoring, and handoff documentation included. If we miss a milestone, that’s our problem.

Scope a build See what's included ↓
FIG. ATypical timeline4–12wks
FIG. BStarting investment$25K+
FIG. CWorking demosWeekly
FIG. DFixed scope and price100%
Sec. 01 — What you get

Working software in your environment. Not a slide deck.

Every Build engagement ships the same set of artifacts to production. No half-working demos, no “we’ll wire up monitoring later.” Production-grade means running on day one of handoff.

DLV-01
Production AI feature(s)

Running in your infrastructure or ours. Tested, instrumented, and reviewed by a senior engineer outside the build team.

Day-one operational
DLV-02
All source code + IP

You own everything. Codebases live in your git, models live in your accounts, no vendor lock-in. We’re builders, not landlords.

Yours forever
DLV-03
Weekly demos in your Slack

Every Friday. Working software, not status updates. You see progress in real time and steer before it’s too late to steer.

Slack-first cadence
DLV-04
Monitoring + alerting

LLM call tracing, error rates, latency, cost dashboards, on-call alerts. Wired up before handoff, not as a “phase 2.”

Datadog · Langfuse · Sentry
DLV-05
Security review

Prompt injection testing, PII handling audit, role-based access, secrets management, rate limiting. Documented threat model.

Threat model + remediation
DLV-06
Handoff documentation

Architecture diagrams, a runbook for every failure mode, prompt library with rationale, evals you can keep running.

Runbook + evals
Sec. 02 — How it works

A 6-week rhythm. Demos every Friday.

Most Build engagements run 4–12 weeks. Here’s the rhythm of a typical 6-week sprint. The cadence holds whether you’re building one feature or three.

Week 01 — Foundation
Architecture lock and technical spike
We finalize the architecture, scaffold the repo, set up CI/CD, and de-risk the highest-uncertainty component first. By Friday demo, you see the riskiest part already working.
Weeks 02–03 — Core build
Feature implementation, end to end
The main feature gets built and wired through every layer: UI, API, model layer, data layer, integrations. By Friday of week 3, you have something a real user can actually use.
Weeks 04–05 — Polish + harden
Eval suite, edge cases, security review
We build the eval suite (golden examples, regression tests, cost tracking), grind through edge cases, and run a security review. Real-world data gets thrown at it.
Week 06 — Production handoff
Deployment, monitoring, runbook, training
Production deployment, monitoring and alerting wired up, runbook written, two training sessions with your team. Final demo, code transfer. By Friday, your team is operating it without us.
Sec. 03 — What we build with

Modern stack. No hammer in search of a nail.

We pick the stack based on what ships best for your use case. Here's the default toolkit, but every choice is a real decision, not a habit.

Models
Claude (default)GPT-4 / GPT-4oGemini (long context)Open-source (self-host)
Retrieval
PineconepgvectorTurbopufferWeaviate
Frontend
Next.jsReactVue 3shadcn/ui
Backend
TypeScript / NodePython / Django / FastAPIPostgres + RedisCelery
Infra
VercelAWS / GCPDockerOn-prem when required
Observability
LangfuseDatadogSentryBraintrust (evals)
Voice
VapiRetellBlandDeepgram (STT)
Dev tooling
CursorClaude CodeGitHub Actions
Sec. 04 — What it costs

Fixed price. Fixed scope. No hourly billing.

Every Build engagement is priced before kickoff. You know exactly what you’re paying. No scope creep, no rate-sheet surprises.

BUI-01Sprint
4–6 weeks · single feature · one workflow. Best for shipping a focused AI feature into existing software.
from $25K
BUI-02Standard
8–12 weeks · multi-component · production-grade. Best when you’re building something net-new with multiple integrations.
from $75K
BUI-03Programme
12+ weeks · multi-team · staged rollout. Best for replacing a core system or building a platform internal teams will use.
from $175K
Milestone-based billing: pay as we ship
Miss a milestone, we eat the difference
You own all source code + IP
Sec. 05 — Recent engagements

Real shipped systems. Real results.

Anonymized because the work was delivered under NDA. The numbers are real; we only publish metrics we can stand behind.

Industrial Distribution · 200+ suppliers
Zero-touch order pipeline across 200+ supplier portals.

Django platform with five integrated systems: real-time order management, supplier scrapers with vendor-specific parsers, instant failure detection with auto-rerouting, milestone customer notifications, and automated payment reconciliation. 2.5× throughput in a 16-week build. Read the full case →

HVAC Distribution · B2B equipment
AI-powered CRM for quoting and order management.

Quoting lived in inboxes and spreadsheets. We built an internal CRM with AI in the quoting and order flow. It worked well enough that the client later opened it up to their own customers. Vue 3, TypeScript, Directus, deployed on-prem.

Sec. 06 — Common questions

Things people actually ask.

What does “production-grade” actually include?
Eval suite (golden examples + regression tests), monitoring (LLM tracing, error rates, latency, cost), alerting on the things that should never silently break, security review (prompt injection, PII, access control, rate limiting), a runbook for every failure mode, and full handoff docs your team can operate from. We don’t ship something that “works on Tuesday” and call it done.
What if requirements change mid-build?
Small changes within scope are absorbed without re-pricing; we plan for normal evolution. Bigger changes get a written change order that adjusts scope, timeline, or both. We’d rather pause for a 30-minute conversation than ship the wrong thing on schedule.
Who owns the code, models, and data?
You do. Code lives in your git from week one. Model API keys live in your accounts. We don’t take exports and we don’t keep “our copy” of anything. When the engagement ends, you don’t lose anything by us walking away.
Can you work alongside our existing engineering team?
Yes, many of our Build engagements run that way. Your engineers get pulled into design reviews, code reviews, and the eval suite. By handoff they’re not learning the system from a doc; they helped build it.
What if AI isn’t actually the right answer?
We’ll tell you. If the cleaner solution is deterministic code, a 30-line workflow automation, or a cheaper tool that already exists, we say so. We’ve turned away Build engagements that should have been a 2-day script.
Do we need to do Discover first?
No. If you already know what you want to build, we’ll scope it and start. Discover is for choosing between options, not a prerequisite.

Ready to ship in weeks?

Book a 30-min call. We’ll talk about what you want to build, scope it, give you a fixed price, and start the week after if it’s a fit.

Book a 30-min Call
If it's not a fit, we'll say so on the call instead of stringing you along.
← 01 · Previous
Discover — decide what’s worth building