Skip to content
POET
[01] Build

Build the AI systems your business actually runs on.

Production AI engagements for operators. We design, build, and operate the agents, internal tools, and custom platforms that move real work off your team — wired into the data and tools you already run on.

[01] Agentic workflows

Move work off your team without losing the audit trail.

We build agents that operate inside your stack — your CRM, your ticketing system, your data warehouse — with explicit policies, scoped permissions, and full observability. Every action is logged. Every decision is reviewable.

  • Multi-step task agents on your tools (Salesforce, Linear, HubSpot, Zendesk, custom)
  • Human-in-the-loop checkpoints for high-stakes actions
  • Policy-based access scoping and audit logging
  • Model evals tied to your real outcomes, not benchmarks
[02] Internal platforms

Custom admin and ops tools your team will actually use.

When off-the-shelf tools cost more in workarounds than they save, we build the platform. Fast, opinionated, designed around your team's actual workflow — not a generic SaaS persona.

  • Admin and back-office platforms
  • Internal RAG and retrieval pipelines on your private data
  • Workflow tools that replace 4-tab juggling
  • Built with the same quality bar as our public product work
[03] Stack we operate on

Frontier providers, production tooling, opinionated defaults.

We pick the right model for the job and the right scaffolding around it. We've shipped on Anthropic, OpenAI, Google, and open-weights — and we keep our opinions sharp on cost, latency, and capability tradeoffs.

  • Anthropic, OpenAI, Vercel AI SDK, LangChain only where it earns its weight
  • Postgres + pgvector for retrieval at our scale
  • Inngest / Trigger.dev for durable workflows
  • Observability via Helicone, Langfuse, or your own
[FAQ] Common questions

How Build engagements actually run.

Do you build the model or use existing providers?

We use frontier providers (Anthropic, OpenAI, Google). Fine-tuning and custom training only when the case is clear — most production AI value comes from system design around capable models, not bespoke training.

Can the agents act inside our private systems?

Yes. We integrate via your existing APIs, OAuth, or service accounts with the same access scoping your team already enforces. Nothing leaves your perimeter unless you explicitly enable it.

Who owns the code?

You do. We ship the repository, the infrastructure-as-code, and the runbooks. Most clients keep us on a retainer for evolution; some take it fully in-house.