Former molecular biologist (RNA splicing, spliceosome mechanics) turned AI-native developer. I enjoy thinking about system design, architecture, data flow, and the bigger picture. I learn by building, which is why I keep creating projects, whether small or large (SaaS).
- 🧠 AI implementations across domains — agents, multi-agent systems, LLM pipelines
- 🏗️ Webapp engineering — FastAPI, JavaScript, Node.js, async everything
- ♟️ Chess — developed my own spatial/strategic framework independently of established theory
- 🧬 Computational biology — where it all started
I run a Signal Coding workflow — every git diff passes through an Advisor Council of AI agents (Security, QA, Senior Dev, CTO, PM) before deployment. Treating AI outputs as experimental results requiring controls, not oracles to trust blindly. From the lab bench to the terminal.
→ Writing about it on Medium @yusupr
Before a single line of code is written, the feature lives as a written specification. Each spec defines the exact deliverable, acceptance criteria, files touched, and exit conditions. The agent implements against the spec; a human reviews the diff and merges only when all gates pass.
spec → branch → implement → test → multi-agent review → human merge
↑ ↓
└──────────────── never skip steps ─────────────────┘
Hard rules:
NEVERcommit to main directlyNEVERstart the next spec until the current one is merged and greenNEVERmerge — human merges after reviewNEVERskip the gate: unit tests + E2E + lint
Active example: LearnX-CLI — a .md → LLM curriculum → TTS audio → MP4 video pipeline built entirely spec-by-spec across v0 → v4, with 235 tests and a Docker container sandbox replacing git-branch-only sandboxing.
NAD Metabolism and Proteomic Profile in a Yeast Model Expressing a Neurotoxic polyQ Protein: Effect of Phenolics from Extra-virgin Olive Oil Vincenzetti S, Rozimemet Y, et al. — preprints.org, 2024 →