Common Questions

Only if you choose. We deploy on Bedrock, Azure OpenAI, self-hosted vLLM/Ollama, or a hybrid. SOC2-, HIPAA-, and SEC-constrained workloads default to private deployment. The decision is yours, written into the SOW, before any data moves.

Every engagement ships with a custom eval suite - graded test cases that exercise the system end-to-end, run on a schedule, with metrics for faithfulness, precision@k, latency, and hallucination rate. The harness lives in the repo, runs in CI, and fails the build when a regression slips in.

Python or TypeScript backends; Next.js or Astro frontends; Postgres; Vercel, Cloudflare, or AWS for hosting; pgvector / Pinecone / Weaviate when retrieval is involved; LangSmith or Langfuse for traces. Stack choices are pragmatic, not religious - we pick what fits your team.

Yes - we offer optional retainers ($8K-$25K/month) for ongoing eval coverage, model updates, observability, and iteration. We can also hand off cleanly to your team or a future hire - that's the default. We do not lock you in. If you want to bring it in-house at month six, we'll spend the last two weeks on a clean handover.

Hours saved per week per team member on the workflows we touch. Error rate before and after. Adoption rate - how many people on the team are still using the new workflow 30 days in. And where applicable, revenue or cost impact.

We measure baseline before we change anything, so the after-numbers actually mean something. We won't show you a dashboard of "AI sessions per week" - that's adoption theater.

That's most of our incoming calls. Stalled pilots usually fail for one of three reasons:

• The wrong workflow was picked (high-visibility but low actual leverage)
• The tool was too generic (an enterprise license that no one customizes)
• Nobody owned getting it adopted (training happened, then nothing)

We diagnose which one it is in week one, then fix it. We are explicitly comfortable inheriting messy situations.

Two 30-minute interviews per person involved during the mapping phase, plus optional weekly office-hour sessions during the pilot phase. We do not require all-hands trainings or week-long offsites.

Adults learn AI fastest by using it on their actual work - not in a sandbox, not in a generic course. Our model is built around that fact.

Yes - non-technical teams are the typical advisory client. Sales, marketing, finance, operations, and customer success are where most of the day-to-day AI leverage actually lives.

We translate technical possibilities into workflows the team will actually use. If a workflow truly needs custom code, we'll tell you, and our engineering side can build it. Most don't.

You keep the workflows, the playbooks, the training materials, and (if we built tooling) the code. We offer optional check-ins quarterly.

Most clients move to a light retainer after the initial engagement; some bring it fully in-house, which we are happy to support. We do not lock you in. The point is for it to keep working without us.