Case studies
Case studies: how I built it
Architecture, the decisions behind it, and the bugs that cost me a night — for the projects I can document from real source code.
An automated YouTube channel for children: animation, voices, quality checks and approvals, orchestrated with n8n
A children's English channel where each episode is written, voiced, animated, checked and uploaded by a pipeline, and a human approves every script and video in Telegram before anything is published.
Read the case studyBuilding a multi-tenant RAG chatbot platform with FastAPI, ChromaDB and a four-level LLM fallback
A self-hosted SaaS where each client gets an AI assistant that answers only from its own website and documents. Architecture, the RAG pipeline, and six production bugs worth remembering.
Read the case studyA virtual try-on mobile app on free GPUs: React Native, FastAPI and diffusion models with graceful fallbacks
Photo of you + photo of a garment → a photoreal try-on image, then a rotatable 3D model. Built entirely on free hosted GPUs, which forced every call to have a fallback.
Read the case studyAppFullSport: a full-stack fitness app with a local LLM coach, React Native, FastAPI and PostgreSQL
Workout tracking with automatic personal records, nutrition with AI food-photo analysis, a social layer and an LLM coach that runs on a CPU-only server. Including the native Android crash that took four attempts to kill.
Read the case studyA natural-language-to-SQL chatbot for MS SQL Server: schema pruning, two LLMs and the guard rails it still needs
Ask a database a question in plain English and get a sentence back. How the schema is pruned before it reaches the model, why two different LLMs are used, and the guard rails a production version needs.
Read the case studyServer monitoring that talks to me: a Telegram alerting bot for SSH logins, service health and tampered repositories
The monitoring I run on my own servers: alerts for never-seen SSH logins, revoked keys, authorizedkeys changes and services down, plus a watcher that scans my GitHub repos for injected malware. Two generations — Bash + cron in production, a Python rewrite for others.
Read the case study