Software Engineer | Systems Architecture | Open Source
Architecting reliable software systems through rigorous engineering. Implementing clean architecture, performance optimization, and strict testing to ensure long-term maintainability and system reliability.
A Multimodal AI Engine that enables users to "chat" with video content by analyzing both audio speech and visual screen context (OCR).
Unlike simple text-wrappers, this project is engineered for depth and reliability:
- 👁️ Visual RAG: Implements an OCR pipeline to "read" code and slides from video frames, capturing context that is shown but not spoken.
- 🧠 Hybrid Search: Uses Qdrant to combine semantic embeddings with keyword search (BM25) for precise technical retrieval.
- ⚡ Performance: Achieved 0.14 RTF on CPU using custom int8 quantization.
- 🛡️ Engineering: Fully Dockerized ETL pipeline with strict type safety guarantees.
I believe in improving the tools I use daily.
- roboflow/supervision: Led the strict typing migration for the
metricsmodule (+700 lines refactored) and ensured full compatibility with NumPy 2.0. Refactored core calculation logic inMeanAveragePrecisionto align the library with production-grade engineering standards. - qdrant/qdrant-client: Enhanced Developer Experience (DX) by standardizing type hints across the client interface, streamlining hybrid search (Dense + Sparse) integration for gRPC and HTTP users.
- SYSTRAN/faster-whisper: Proposing architectural patterns for Dynamic Batching and request queuing to solve latency bottlenecks in high-concurrency GPU environments.
- pyDeprecate: Contributed to core design discussions on decorator deprecation lifecycles, advocating for "Fail Fast" import-time warnings to prevent silent technical debt in production systems.
- Core & Backend: Python 3.12 (Strict Typing), Docker, FastAPI/Streamlit.
- AI & Data: OpenAI, Faster-Whisper, RapidOCR, Sentence-Transformers.
- Vector Search: Qdrant (Hybrid Search: Dense + Sparse).
- Quality & DevOps: Mypy, Ruff, Pre-commit hooks, GitHub Actions.