Lead software engineer · backend and applied AI systems
Backend services for document processing, data validation, model inference, and reliable operations.
Website • Engineering notes • LinkedIn
- ConTrust Suite is a multi-tenant construction reporting platform.
- Its backend handles PDF ingestion, OCR, extraction, validation, SharePoint sync, reporting, billing, and AI briefs and questions.
- The system keeps source files, raw values, validation flags, and evidence for review.
- The OCR service processes Arabic and English financial statements. It classifies pages, extracts values, and checks document structure.
- The Gemini Nano project studies Chrome's on-device model from a measured 3.98 GB
weights.binfile. - The project maps the file layout and implements analysis and inference code in Python.
- Findings are labeled as reverse-engineering results. They are not treated as vendor documentation.
- Common work includes API design, background jobs, database services, external data sources, retries, observability, and failure recovery.
- The main languages are Python, TypeScript, Rust, Go, and Solidity.
- QuakeReunite : search API and maps for people and aid centers after the June 24, 2026 La Guaira earthquake. The pipeline uses FastAPI, SQLite, Mistral OCR, and Gemma 4 on Cerebras. A maintenance loop refreshes the index every ten minutes.
- sqlbench-harness : evaluation harness for SQL generated by language models. It records result accuracy, execution errors, token use, provider cost, and fairness class. It separates fair runs from diagnostic runs.
- WDK Browser Extension Starter : Chrome and Brave wallet extension starter with an encrypted local vault, background-worker key custody, origin-scoped dApp access, and transaction review.
- Aave V3 Data : daily Aave V3 reserve data for 13 networks. GitHub Actions publishes static JSON and HTML without an application server or API keys.
- Colab Inference : local OpenAI-compatible API for language models running on Google Colab GPUs.
| Area | Tools and systems |
|---|---|
| Languages | Python, TypeScript, Rust, Go, Solidity |
| Applied AI | OCR, document intelligence, retrieval, agents, multimodal extraction, evaluation |
| Backend and data | FastAPI, Node.js, NestJS, PostgreSQL, MongoDB, SQLite, WebSockets |
| Inference | vLLM, SGLang, TensorRT-LLM, CUDA, NCCL |
| Cryptography | Foundry, Hardhat, Ethers.js, BLS12-381, Merkle proofs, WASM |
| Infrastructure | Docker, Google Cloud, GitHub Actions, background workers |
The work is measured with source evidence, reproducible tests, and deterministic validation.