Technical Work

Building GPT-2 scale models from scratch to deeply understand transformer architectures, attention mechanisms, and training dynamics

Developing MCP servers for enhanced LLM tool use and context management, enabling better integration between language models and external systems

Exploring chain-of-thought reasoning, RLHF techniques, and implementation of reasoning capabilities in smaller language models

Multimodal AI Hackathon at Weights & Biases HQ. Built an AI assistant that analyzes product screenshots, creates instant knowledge bases, and generates purchase insights using SambaNova vision models and Google Gemini

LangChain AGI Hackathon (2023). Built an advanced AI assistant that integrates vision-language models (VLMs), large language models (LLMs), and MultiON AI to perform real-world tasks through browser automation

Integrated Maxine SDK with Triton Inference Server for large-scale inference with efficient batch processing and optimized computational performance

Biorobotics Lab - Developed HMM-based regression system for skill evaluation with integrated pose estimation pipeline

Developed complete simulation environment for multi-robot coordination in underground environments using ROS and Gazebo

Formulated and implemented bundle adjustment optimization in Ceres Solver for offline localization with AprilTags

Implemented imitation learning for traffic agent modeling in CARLA with DeepLabV3+ segmentation and multi-object tracking

Implemented classical computer vision algorithms for 2D image registration using CUDA/C++ for real-time performance