About

I got into AI because the problems never get boring.

I'm Matt, Principal AI Engineer at Lucanet. I work with startups and SMEs who have an AI problem and no one to build it.

40%
Ticket deflection LISA · Lucanet
57%
Flood reduction ARROW
70%
Cost cut vs. SaaS LLM Enrichment
10x
Research speedup Earth Security

I like building things. SaaS, infrastructure, energy, ecology, go-to-market. Different data, different constraints, different set of people telling me why the last AI project didn't work.

The work asks for more than just code. Data science, software engineering, understanding the business well enough to build the right thing. Most roles ask for one of these. This one needs all of them.

AI moves fast. What I could build a year ago looks nothing like what I can build now. The models improve, the tooling shifts, and every few months there's a better way to do something that was already working. The hard part isn't building it. It's knowing which approach will still make sense in six months.

See recent projects
Tech Stack

AI / LLM

LangGraph RAG MCP OpenAI Claude Pydantic AI LiteLLM Langfuse CrewAI

Backend

Python TypeScript FastAPI PostgreSQL SQLAlchemy Vector DBs

Frontend

React Next.js Streamlit Chainlit

Infrastructure

AWS Azure Terraform Docker GitHub Actions N8N

Data

Pandas Pydantic dbt Airflow Accrue
How I work

A few things that matter to me.

Every project is different, but these stay consistent.

Start with something real

I like to get a working prototype in front of you in a couple of weeks. Your real data, your actual problem.

One point of contact

I scope it, build it, and explain it. Same person on every call.

Show the work early

Every check-in has a demo. If something's not right, we catch it early.

You own everything

Code, docs, deployment config. Your team can run it, modify it, or bring someone else in. No lock-in.

Honest about fit

I don't have a platform to sell. If a simpler tool solves the problem, I'll say so.

Start with something real

I like to get a working prototype in front of you in a couple of weeks. Your real data, your actual problem.

One point of contact

I scope it, build it, and explain it. Same person on every call.

Show the work early

Every check-in has a demo. If something's not right, we catch it early.

You own everything

Code, docs, deployment config. Your team can run it, modify it, or bring someone else in. No lock-in.

Honest about fit

I don't have a platform to sell. If a simpler tool solves the problem, I'll say so.

I've killed more projects in discovery than I've shipped. That's the point — the expensive mistake is building the wrong thing, not spending a week figuring out you shouldn't.

Matt House
~
$ cat matt.yml
name: Matt House
role: AI Engineer
location: wherever the food is best 🍜
$ pip install accrue
✓ enrichment engine installed
$ claude "what's matt up to?"
building an HR support agent — almost there

Got a problem worth solving?

30 minutes. No pitch, no slides — just the problem and whether I can help.