I turn messy, real company processes (steel pricing, contract registers, H&S programmes, supplier email) into software that people use every day: Next.js + PostgreSQL on the web side, Python for data, ML and LLM pipelines.
- 🏭 Three of my apps run in production or internal testing at a large industrial company
- 🧠 Currently learning RAG pipelines in Python (retrieval, chunking, evaluation)
- 🤝 Open to B2B collaboration: internal tools, process automation, local/private LLM integrations
- 📫 LinkedIn · matitrampolin@gmail.com
| Project | What it is | Stack |
|---|---|---|
| SteelQuote | Pricing and offer platform for steel sales teams: real-time pricing for 6 steel types, configurable approval flows and validation rules, role-based visibility, PDF/Excel export, analytics. v2.0 live in production. | Next.js 14 · TypeScript · PostgreSQL |
| AMSafe | Gamified H&S (BHP) PWA for a production warehouse: daily quizzes, missions, points and ranks, worker → leader → manager hierarchy, push notifications. In production. | Next.js · Prisma · PostgreSQL · Web Push · S3 |
| CRU2026 | Rebuild of a legacy CakePHP contract registry: server-side read authorization, SharePoint storage through Microsoft Graph, migration of ~20k legacy records. | Next.js 14 · Prisma · PostgreSQL · MS Graph |
| Logistics Network Map | Turns shipment exports into an interactive route map with tonnage-weighted routes, hubs and top customers, in one offline HTML file. | Python · pandas · OSRM |
| Project | What it is | Stack |
|---|---|---|
| AI_Mail_Arbitrium | Classifies supplier email replies with a local LLM: JSON-Schema constrained decoding, verbatim evidence grounding and a deterministic human-review queue. Nothing leaves the machine. 217 tests. | Python · LM Studio · Qwen3 · Outlook COM |
| Synthara (private) | Multi-persona AI content platform: FastAPI + Celery workers, ComfyUI image/video generation, n8n scheduling and a Next.js human-in-the-loop review dashboard. | FastAPI · Celery · Redis · PostgreSQL · ComfyUI · n8n · Docker |
| LunaCore HUD | Electron dashboard for the Claude Code CLI: context and usage telemetry, session tabs, and parallel AI workers in git worktrees that finish as PRs. | Electron · JavaScript · Python |
| Project | What it is | Stack |
|---|---|---|
| money_printer_turbo | Research harness for US-equity strategies with pre-registered kill gates, a deflated Sharpe ratio and fill-realism checks on real quotes. Three theses honestly killed, and $1,759 of data purchases avoided by gates. | NautilusTrader · CatBoost · Databento |
| crypto-regime-trader | Binance bot: pattern + market-regime engine, LLM news sentiment, Kelly sizing, scale-out exits, walk-forward backtests, pytest suite. Testnet-first. | Python · Flask · pandas |
Older experiments: news & analyzer tools · newsletter tools
- Frontend: TypeScript · React · Next.js (App Router) · Tailwind CSS · shadcn/ui · PWA
- Backend: Node.js · Prisma · PostgreSQL · NextAuth · FastAPI · Celery · Redis · REST
- Data / ML: Python · pandas · NumPy · CatBoost · NautilusTrader · walk-forward / CPCV validation
- LLM: local models (LM Studio, Ollama, Qwen) · structured output / JSON Schema · prompt pipelines · ComfyUI · learning: RAG
- Integrations & ops: Microsoft Graph / SharePoint · Outlook COM · n8n · Docker · Vercel · Electron · Git
- 🔎 RAG pipelines in Python: retrieval, chunking strategies, evaluating answer quality
- 🧠 Applied ML and LLMs beyond demos: private, local models doing real business work
- 📈 Algorithmic decision systems and finance, validated honestly (walk-forward, kill gates)
- 🏗️ Backend architecture: authorization, data migrations, systems that are easy to run
- 🌐 Networking fundamentals: TCP/IP, traffic analysis (Wireshark, tcpdump)
I like building systems designed for an AI-focused future, grounded in solid understanding. I believe AI is often misunderstood, but it has the potential to fundamentally change the way we live and interact with the world.