Lead Software Engineer · AI & LLM Systems · 12+ years
Website · LinkedIn · X · Email
Lead engineer with 12+ years shipping production systems. For the last three years I've been building production GenAI and agentic systems end to end: multi-agent platforms, LLM orchestration, and the Kubernetes infrastructure that runs them.
📍 India · Open to Senior / Lead roles — relocation or remote
| Project | What it does | Stack |
|---|---|---|
| Video Factory | Submit an idea and a 7-stage pipeline generates transcript, audio, images, subtitles, and the final video — with YouTube upload built in | Next.js · Cloud Run · YouTube API |
| google-health-sync | Self-hosted personal health assistant that syncs Fitbit/Pixel Watch data from the Google Health API, with a chat assistant that remembers you, a diet planner, and a weekly report | Python · LiteLLM · Docker · Raspberry Pi |
| AI Learner | Turns any PDF, page, or video into a daily reinforcement-based study loop delivered over Discord with AI feedback | LangChain · Python · Next.js |
| NewsFlow | Reel-style news reader that turns RSS feeds into swipeable, image-rich cards | Next.js · RSS · Mobile-first |
| TravellerTip | Travel hub covering visa requirements, political relations, common scams, and destinations | Next.js · Postgres |
| Chess Arena | Benchmark engine that runs LLMs against each other in chess matches and knockout tournaments, logging legality, retries, tokens, and cost for every move | Next.js · Python · Postgres · Cloud Run |
- research-papers — 100 landmark deep-learning papers, complete, reimplemented from scratch: Word2Vec, Dropout, Adam, VGGNet, GoogLeNet, Batch Norm, ResNet, seq2seq, Neural Turing Machines, Transformer, VAE, and more. Each paper gets a runnable
solution.ipynb, an architecture breakdown, and an Open in Kaggle notebook linked back to the original arXiv publication. - llm-research-papers — Season 2: 100 LLM-era papers from RMSNorm (2019) to 2026, reimplemented from scratch (in progress).
- jev-crash-course — an 11-level crash course on Jev, TypeSafe AI's System One decision model. Runnable examples against the real API, plus a capstone ticket-triage project with unit tests and evals. Provider-agnostic via LiteLLM.
- LangChain — added
page_labelmetadata toPyPDFLoaderso PDFs with front matter map to real page numbers (merged) - LangFlow — fixed duplicate primary-key indexes in database migrations (merged)
- 9router — added Ollama support for System One decision models (clef, clef-flash), routed through the
/v1/systemoneendpoint for cloud and local (merged) - ai-scan-job — LLM-powered GitHub PR reviewer that posts review comments straight into your workflow
- py-cron — lightweight image for running Python scripts on a cron schedule
Resume, experience, and contact: devnadeem.com




