The Engineer's Codex · MMXXVI
Full-Stack Software Engineer & Applied ML · 6 Years Exp · MSc CS
Full-Stack & Applied ML Engineer with 6 years of commercial experience. Deploying production AI in enterprise, building scalable web applications, and running software house CodeWorks (codeworks-it.pl).
Production systems built with an instrument-maker's care — Python & FastAPI below the waterline, React & Next.js above it, applied machine learning throughout.
Enterprise AI (Services Platform) · MSc in Computer Science · Founder @ CodeWorks
§ I — Selected work
This folio records two kinds of work: commercial deployments engineered for clients and kept in production, and mechanisms from my own workshop — built on personal initiative to probe the frontiers of local AI and systems engineering. Each plate is a working instrument, reproduced here at reduced scale, with figures drawn from the originals.
Commercial Deployments
Fig. 1 — The AI intelligence engine in service procurement
An enterprise-grade platform integrating language models and semantic evaluation pipelines directly into service discovery and purchasing flows. The system automates inquiry routing, validates matching criteria in real time, and interfaces directly with core business operations.
Enterprise scale · real-time inference · automated routing
Fig. 2 — The three-stage QA circuit
A batch machine-translation pipeline that carried roughly a million catalogue rows across languages. CTranslate2 supplies the throughput; a three-stage QA circuit — rule checks, back-translation, human sampling — decides what ships. Postgres keeps the ledger, and every failure is resumable.
≈1,000,000 rows · 3-stage QA · resumable batches
Fig. 3 — Subject lifted, ground restored
A product-photography pipeline that lifts subjects with a fine-tuned YOLO segmentation model and restores the vacated background with LaMa inpainting. Replaces painstaking manual retouching with automated batch processing — over 50,000 images brought to catalogue standard in a fraction of the time.
≈50,000 images · YOLOv8-seg masks · LaMa infill
Fig. 4 — The reference library, wired
Retrieval-augmented answering over a private corpus of 2,000 documents — LangGraph orchestrates the flow, and llama.cpp runs local CPU inference on modest hardware. Answers cite their source leaves, and nothing leaves the infrastructure.
≈2,000 documents · local inference · CPU-honest
The Workshop
Fig. 5 — The virtual drive, indexed
A privacy-first desktop agent in Rust that mounts a virtual semantic drive (X:) over WinFSP and turns static folders into a queryable knowledge network. Keyword and vector retrieval fused by reciprocal rank; OCR, visual and voice search run entirely offline. Its agent mode proposes bulk file operations — a human approves them before anything touches disk.
100% local · virtual drive X: · human-in-the-loop
View the source ↗Fig. 6 — Candidates at the gates
A deterministic engine that redacts AI-drafted Polish legal prose — no generative model in the loop. Rule-generated candidates pass layered validators that guard normativity, parties, amounts and dates; Stanza syntax and a semantic filter stand as optional gates, and every decision is logged to a JSON report.
High precision profile · validator gates · legal entity guard
View the source ↗Fig. 7 — Specimens of the system
A design system of warm paper, aubergine ink and sealing-wax orange: tokens, base styles, components and a live tweaks panel, with demos for desktop, mobile and dashboard. The very system this folio is set in.
Tokens · components · tweaks panel · this very folio
View the source ↗§ II — Cursus honorum
Six years drawn as an orrery — formation at the centre, each engagement in its own orbit. Drag to turn the instrument; select a body to read its entry.
2024 — present
APIs, search algorithms and PostgreSQL on the platform side; Python, LangGraph, LangChain and Ollama bringing local LLM inference into production. The current commission.
Backend Developer — TME2024 — present
APIs, search algorithms and PostgreSQL on the platform side; Python, LangGraph, LangChain and Ollama bringing local LLM inference into production. The current commission.
TypeScript · PostgreSQL · Python · LangGraph · Ollama
Full-Stack Developer — Ecohedge2023 — 2024
Data visualisation in React and React-Table; Auth0, Mailgun, Codat and OpenAI integrations over MongoDB — the first commercial brush with LLM APIs.
Next.js · Jotai · MongoDB · Auth0 · OpenAI
React Developer — WEUPCODE2021 — 2023
Sixteen months of steady React craftsmanship: TypeScript, Redux and Tailwind against REST services, shipping features week over week.
TypeScript · Redux · Tailwind · REST
Junior Full-Stack Developer — Softwarebay2021
React and Next.js at the front, a Laravel back-end maintained in PHP — the first full-stack commission.
React · Next.js · Laravel · PHP
Junior Web Developer — KS Sport2020 — 2021
First post. JavaScript and PHP solutions designed and built from scratch, with an SQL database kept in good order.
JavaScript · PHP · SQL
BEng & MSc, Computer Science2017 — 2023
Bachelor of Engineering at the Polish Naval Academy, Gdynia (2017–2021) — web programming and DevOps. Master of Science at WSB Gdańsk (2021–2023) — front-end specialisation, completed while already in service.
Web · DevOps · Front-end
§ III — Commercial Engagement
Looking for a production AI deployment or complex automation? I run software house CodeWorks (codeworks-it.pl) focused on AI deployments, process automation, and B2B systems for e-commerce and wholesale.
Visit codeworks-it.pl ↗§ IV — The archive
A card catalogue over the engineer's records. Retrieval runs entirely within this page — no server is consulted.
Records of the engineer · Drawer no. II
The drawer stands open.
Ask about deployment, pipelines, retrieval, or testing.
§ V — Index of instruments
§ VI — Correspondence
Letters, commissions, and curious questions are received at any hour. Replies dispatched within the day, wax permitting.
Poland · UTC+2 · Open to contract & remote