Hello, I'm
Building intelligent systems that bridge the gap between complex algorithms and real-world utility. Focused on scalable deep learning and precision engineering.
I don't just train models; I build resilient AI systems. My approach is rooted in understanding the mathematical foundations of machine learning while engineering practical, scalable solutions.
Whether optimizing inference latency or architecting distributed training pipelines, I prioritize clean code and reproducibility. Technology is a tool, but rigorous engineering is the craft.
Problem: Model latency was too high for real-time edge deployment.
Engineered a custom inference server using C++ and ONNX Runtime, achieving a 40x speedup over the Python baseline. Implemented dynamic batching and request queuing to maximize GPU throughput under high load.
Problem: GPS-denied navigation in complex indoor environments.
Developed a SLAM-based navigation stack using ROS2 and LiDAR data. Integrated a Reinforcement Learning agent (PPO) for obstacle avoidance, training it in Unity for 10M steps before sim-to-real transfer.
Problem: Hallucinations in LLM responses for legal queries.
Built a Retrieval-Augmented Generation pipeline using LangChain and Pinecone. Implemented a custom re-ranking mechanism and citation validation layer to ensure high accuracy and traceability of sources.
Exploring quantization techniques and hardware-aware pruning to run BERT-class models on mobile hardware without significant accuracy loss.
Why monolithic architectures might be a better choice for early-stage ML infrastructure, and when to actually migrate.
A deep dive into teacher-student networks and how to effectively compress large models for production environments.
Currently open to new opportunities in AI/ML engineering. If you're building something challenging, I'd love to hear about it.
