SEC 01 — AI INTEGRATION · SHT 05
RAG over your data, agents and copilots, semantic search, and fine-tuned models — inside your existing stack, with evals and monitoring.
IN YOUR STACK · EVALS INCLUDED · MONITORED IN PROD
Scope your integrationSee the AI Support Agent blueprintSEC 02 — WHO IT'S FOR
THIS IS FOR YOU IF
NOT A FIT IF
SEC 03 — WHAT'S INCLUDED
Scoped per feature set. Every item ships production-grade or not at all.
D1
Ingestion, chunking, embeddings, vector search, and grounded answers with citations — so the model answers from your data.
D2
Multi-step, tool-using, permission-aware agents, and copilots beside your users' workflows. Explicit boundaries, human-in-the-loop where stakes require it.
D3
Meaning, not keywords: embeddings, vector indexes, and hybrid ranking tuned against your real query logs.
D4
Where prompting and RAG top out: your domain language, output formats, or cost/latency targets. Training pipeline documented and owned by you.
D5
Eval suites for accuracy, groundedness, and regressions, run in CI. Production monitoring for latency, cost, and answer quality.
D6
Built inside your repo, your infrastructure, your CI/CD, your review process. No parallel stack, no dependency on us to deploy.
SEC 04 — HOW IT RUNS
The five-step Desmok AI process applied to an existing codebase. Full detail on the process page.
SEC 05 — STACK
Chosen per project for longevity and running cost. Documented. No proprietary lock-in.
MODELS
AI TOOLING
BACKEND & DATA
INFRA
The default is your stack. We integrate with what you run — your languages, your cloud, your CI — and add only what the AI layer genuinely needs. Model choice is a cost/quality/latency decision made in Scope, with evals to prove it.
SEC 06 — RELATED BLUEPRINT
BLUEPRINT — AI SUPPORT AGENT
Reference buildA customer-facing agent trained on your docs and policies. Answers on web and WhatsApp, hands off to a human when it should, and logs every conversation. RAG, vector search, and handoff logic — the same building blocks most AI Integration work is made of, opened up.
Stack
SEC 07 — ENGAGEMENT & PRICING
Integration work runs two ways. A fixed-scope project fits a defined feature set: Scope produces a fixed quote with milestone billing, evals define "done", and the engagement ends with a clean handover into your team. A monthly retainer — or a dedicated team — fits roadmaps where AI capacity is ongoing: standing senior engineers at a predictable monthly cost, scaling up or down month to month. Either way the ground rules hold: weekly demos, direct access to the people writing the code, and your IP — including models, prompts, eval suites, and training data pipelines — always yours.
MODEL 03
Embedded engineers and designers working as part of your team. Scale up or down monthly.
Best for: when the work is continuous and yours to direct
SEC 08 — FAQ
Yes — that's the defining constraint of this offer. Our engineers work in your repo, follow your conventions, go through your code review, and deploy through your CI/CD. Features land as PRs, behind feature flags where sensible. We read your codebase in Discover before proposing architecture.
SEC 09 — START
Bring your stack and what you want AI to do. We scope the integration plan: architecture, models, evals, timeline, price. Yours to keep, whoever builds it.