I turn deep learning research into production systems: edge inference at 500+ cameras, biometric payments in live retail, privacy-first RAG without cloud dependency. Currently AI & Software R&D Manager at a large-scale enterprise.
- Pushing inference further on constrained hardware — quantization, pruning, TensorRT pipelines
- Expanding my offline Enterprise RAG stack (Haystack + Weaviate + local LLMs) with multi-modal capabilities
- Kaggle: competitive ML in audio, NLP, and behavioral analytics
- LangGraph-based multi-agent architectures for enterprise workflows
- Advanced ANNS indexing strategies on vector databases (Milvus, Weaviate)
- TinyML deployment patterns for Jetson-class edge devices
| Year | Achievement |
|---|---|
| 2026 | 🥈 Silver Medal – BirdCLEF+ (Top 1.4%, Solo, 4,084 teams) |
| 2026 | 🥈 Silver Medal – Deep Past Akkadian Translation (Top 2%, Solo, 2,673 teams) |
| 2025 | 🥉 Bronze Medal – MABe Animal Behavior Recognition (Top 7%, Solo) |
| 2020–2023 | 📌 Stanford World's Top 2% Scientists (4 consecutive years) |
Vision → PyTorch · OpenCV · TensorRT · ONNX · NVIDIA Jetson
NLP / GenAI → Hugging Face Transformers · Haystack · LangGraph · Local LLMs
Vector DB → Weaviate · Milvus · OpenSearch
MLOps → Docker · FastAPI · MLflow · MinIO · Git
Cloud → AWS (SageMaker, EC2) · Linux
- Edge AI / TinyML optimization challenges
- Privacy-preserving ML systems (on-prem RAG, federated learning ideas)
- Kaggle competitions in CV, NLP, or time-series domains
- Getting production-grade inference out of limited hardware
- Building RAG systems that never touch an external API
- Going from SOTA paper → filed patent → live product
- LinkedIn: https://www.linkedin.com/in/musa-p-46253815a/
- Kaggle: https://www.kaggle.com/musapeker
- Contact: msapeker@gmail.com
"The gap between a research result and a deployed product is where most of the real engineering happens."