AI Engineer & Full-Stack Developer. I build LLM applications, autonomous agents and production ML systems — and I ship them, rather than leaving them in a notebook.
Currently building AI products and contributing to the open-source AI tooling I use, from Mumbai.
🌐 nazsats.com · 📄 Resume · ✍️ Blog · 💼 LinkedIn · 𝕏 @Nazsatseh
I read the source of libraries I actually use, looking for places where the code quietly disagrees with itself. 8 bugs across 5 libraries, 3 merged.
| Project | What I found |
|---|---|
| langchain | DeepSeek reports prompt-cache usage in a top-level field, while the inherited OpenAI parser reads only a nested one — so cache savings came back absent, with no error. Fixed in both the streaming and non-streaming paths. Merged. |
| qdrant-client | Local mode matched points the real server rejected — four range bounds checked independently, so two values could each satisfy half a range. Merged, shipped in v1.19.0. |
| qdrant-client | Two more local-vs-server divergences: MatchExcept treated an explicit null as a value that differs from everything, and an empty should matched nothing where the server matches everything. Merged. |
| Project | What I found |
|---|---|
| Chroma | add() was the only write operation rejecting a None metadata value, even though the type, the validator and upsert() all accept it. |
| LlamaIndex | run_async_tasks silently swallowed exceptions raised inside async tasks when show_progress=True. |
| LlamaIndex | similarity_top_k=0 returned the entire index instead of nothing — a falsy-zero check inverting the request. |
| Outlines | Every term was wrapped in a capturing group, renumbering the groups inside a user's pattern — so Regex(r"(a)\1") raised "cannot refer to an open group" while the named-group form worked. |
| LangGraph | Checkpoint savers disagree on list(limit=-1): InMemorySaver returns nothing, both SQLite savers return everything. |
Each filed with a reproduction, a minimal fix and regression tests that fail without it.
I write them up too — LangChain Couldn't See DeepSeek's Cache Hits and Two Values, Half a Range Each.
Flight Delay Intelligence — a
containerised PySpark pipeline and calibrated LightGBM model over 1.6M US DOT flight records.
Isotonic calibration cut expected calibration error 2.5×, and the alert threshold comes from
the cost of a missed delay versus a false alarm rather than the 0.5 default. The
write-up covers why weather
turns out to be 7%.
PySpark · LightGBM · MLflow · FastAPI · Streamlit
BloodAI — upload a blood test, get it
explained in human words in under 30 seconds. Flags out-of-range markers and builds a wellness
plan around your actual chemistry. There's a
mobile app too.
Next.js · GPT-4o Vision · Firebase
Dubai AI Broker — a multi-tenant SaaS that
does a real-estate agent's admin for them. Claude with tool-calling, grounded in a real
database so it can never invent a listing.
FastAPI · Claude · PostgreSQL · Next.js
AI Store Builder — one sentence to a
complete storefront in about twenty seconds. Model output is schema-validated data rather than
code, so generated content can never be executed or injected.
Next.js · NestJS · Prisma · OpenAI · Zod
AutoProfit — an AI crypto trading system that
executes live trades on Bybit and scores coins from news and social sentiment with an LLM.
Take-profit ladders, risk controls, paper mode.
Python · FastAPI · ccxt · Next.js
Languages Python · TypeScript · SQL · Solidity
AI OpenAI (GPT-4o, Vision) · Claude · LangChain · LangGraph · RAG · agents & tool-calling · MCP · vector DBs
ML & Data PySpark · LightGBM · scikit-learn · MLflow · pandas · probability calibration · drift detection
Backend FastAPI · Flask · Node.js · NestJS · PostgreSQL · Supabase · Firebase
Frontend Next.js · React · React Native · Tailwind · Framer Motion
Web3 Solidity · Hardhat · EVM · Solana · ERC-20/721/1155
Founded CatCents, a GameFi and DAO platform on Monad — grew it to a 30,000+ member community and raised $50,000+ across NFT launches.
Before code, I wrote — 50+ articles and 20,000+ readers. It's why I still write up every bug I find.
📬 Open to AI engineering roles and freelance work — get in touch