Hoyack Labs
Hoyack Engineers Build
Production AI for Businesses
Hoyack is the engineering team. We pick a stack that holds in production (FastAPI, CrewAI, Python, async workers) and run every project through the same discipline we apply on client engagements. The page below walks through how we work. Anvil of Ideas is one proof of what comes out the other side.
How Hoyack builds
Hoyack is the Team that Builds the Projects
Before the products and the demos, there’s an engineering team and a way of working. Hoyack settled on this stack and this discipline after watching less rigorous setups fail in production. It’s how we ship Anvil, and it’s how we’d ship work for your business.
01 · FRAME
Start from a Concrete API Contract
Each project opens with a payload and a response shape: in Anvil’s case, a business name, an industry, and a target audience in, a preview site out. That contract anchors everything that follows.
02 · BUILD
FastAPI + CrewAI on Python
One backend stack across Labs. We moved off n8n because we wanted full control over agents and access to Python’s AI library ecosystem. FastAPI is the API surface; CrewAI runs the crews.
03 · HARDEN
Structured Output that Holds
Pydantic schemas at the API boundary instead of fragile parser nodes. We saw many false-positive failures elsewhere; typed contracts in FastAPI mean the output either matches or fails loudly.
04 · SHIP
Async Agents Behind a Clean API
Long-running agent runs go through RabbitMQ and workers, so the user-facing API stays responsive. Each Labs product is one team’s responsibility end-to-end.
Proof of work · Live
Anvil of Ideas Demonstrates the Work
Anvil is the Labs product furthest along, the one Valeria calls “in production.” Enter a business name, an industry, and a target audience; the front end posts that payload to the Ultimate Website service, a CrewAI crew runs against it, and a generated demo site comes back rendered inside the Anvil page.
An Example of What the Skills Produce
Anvil shows several Hoyack Labs skills in motion at once: a CrewAI crew running off markdown system instructions, a FastAPI service with a typed payload contract, and a separate Ultimate Website module that another product could just as easily call.
Other Hoyack projects
What else the Same Team is Building
Anvil is one product. Here are the others Hoyack Labs is building with the same skills and the same engineering team. They’re early. The skills are not.
Anvil of Ideas
Website generator. Business name, industry, and target audience in; a CrewAI crew runs against the Ultimate Website service and a demo site comes back, rendered inside the Anvil tab.
Iris
Voice agent service. The backend is in place; we haven’t built the front end yet. Once we do, it slots into the same FastAPI + CrewAI pattern Anvil uses.
Video Generator
An internal video generation app Brandon has been building on the same Python stack. Deployment status is being confirmed; not yet a public product.
On the side · Open source on GitHub
Helper Skills We’ve Packaged Up and Shared
While Hoyack ships client work, our engineers have packaged some of the patterns we use into reusable helpers and dropped them on GitHub. They’re not what runs Anvil; they’re side artifacts our team made portable so you can take them too. Free, MIT-licensed, no Hoyack login needed.
MIT
hoyack / crewai-starter
FastAPI + CrewAI Starter
Opinionated scaffold for the stack we use in Anvil: FastAPI service, CrewAI crew, markdown system instructions, typed Pydantic responses. Clone it, point your agents at a problem.
MIT
hoyack / typed-llm-output
Typed LLM Output Helpers
Pydantic-first response wrappers we wrote after one too many silent parser failures in n8n. Either the LLM output matches the schema or it fails loudly. No false positives.
MIT
hoyack / md-system-prompts
Markdown System Instructions
The pattern we use to keep agent behavior in versioned markdown files instead of text boxes. Includes a small loader, a linter, and a few example crew definitions.
MIT
hoyack / agent-worker-queue
RabbitMQ + Workers for Agent Runs
Reference setup for pushing long agent runs onto a RabbitMQ queue with worker processes, so your user-facing API stays snappy. Boilerplate-free, drop-in for FastAPI services.
Notes
hoyack / n8n-to-fastapi
n8n to FastAPI Migration Notes
The lessons we wrote up after moving our agent workloads off n8n. What broke, what we replaced it with, and the FastAPI patterns that finally held in production.
.
Pattern
hoyack / module-as-service
Generator-As-A-Module Pattern
How Anvil calls into the Ultimate Website service: one product, multiple consumers, clean API contract. Includes the patterns and a small example repo you can fork.
.
We left n8n because the AI nodes had unreliable outputs. Even when the response fit the format, the structured-output parser kept rejecting it. We wanted better control of our agents, so we moved to FastAPI and CrewAI.
Valeria Villanueva · Engineering, Hoyack Labs
Want Our Engineers on Your AI Project?
Hoyack’s engineering team takes on production AI work for businesses. The discipline above is what we’d point at your problem. Anvil is one proof of what comes out; the helpers on GitHub are yours to take.
