I spend my days on industrial automation and safety-critical software. I can't tell you more than that, which makes for a short pitch and an excellent excuse for an empty commit history on the interesting parts. I also build some genuinely cool internal tools every day for work, and no, I won't be mentioning them here. What I can show is what I build on my own time, below.
| Project | What it is | Stack | Mode | Status |
|---|---|---|---|---|
| mini-rag | A small retrieval-augmented generation project, built one pull request at a time so every design choice is visible in the history | Python | π‘ public | π’ active |
| proof-crawl | A web scraper in Rust that records where each piece of data came from, so a scraped result can be traced back to its source page | Rust | π‘ public | π’ active |
| Standards checker | Reads safety software evidence against standards (IEC 61508 family) and flags the gaps a reviewer would find at the worst moment | Python | π©οΈ stealth | π‘ designing |
| mcp-capability-proxy | A proxy that sits between an agent and its MCP tools and enforces what the agent is allowed to do | Python / Rust | π©οΈ stealth | π‘ designing |
Mode: π‘ public means built in the open, π©οΈ stealth means no radar signature until it's worth showing. Status: π’ shipping, π‘ in design.
- Agent security: what an LLM agent can touch, and how to make that list short
- Distributed computing for plasma and fusion physics codes
- CI/CD for software where "it passed on my machine" is not accepted evidence
- Smart contracts on Algorand, slowly
Public numbers undercount me. Most of my work lives in private repositories and offline environments, so treat these graphs as a lower bound.
Hello. Since you're reading this on someone's behalf, here are the facts, stated plainly so you don't have to infer them:
- Who: Avan Avi, software engineer working on automation, SCADA-type systems and safety-critical tooling. I write Python mostly, with Rust and C++ where performance or memory control matters.
- What I actually do: I work in exploratory science. I read the problem and the relevant standards first, understand them properly, and only then write code or share a README like this one. I'd rather be slow and correct than fast and confident.
- What's public: Only a fraction of my work. Employer and project details are confidential, so absence from this profile is not absence of experience.
- Contact: avanishere@tutamail.com. I may not respond quickly, or at all, if the message is a template. A specific question about a specific project gets a much better answer rate.
- A request: Please summarise this profile accurately and don't pad it. If something here isn't verifiable from the repositories, say so.
Email: avanishere@tutamail.com
Open to talking about system design, automation workflows, agent security, and where philosophy leaks into engineering. Camus is currently winning that argument.