CLICK + DRAG ANYWHERE TO LAUNCH A SPACECRAFT
I am a Ph.D. with 5+ years of experience applying AI and machine learning to real-world challenges across finance, healthcare, and engineering. My work spans traditional machine learning, deep learning, reinforcement learning, generative AI, and optimization under uncertainty.
On the generative AI side, I specialize in fine-tuning large language models and architecting advanced RAG and agentic RAG systems. I build autonomous multi-agent workflows with LangChain, LangGraph, and CrewAI, backed by Model Context Protocol integrations for tool use, API orchestration, and human-in-the-loop oversight.
My doctoral research at Carleton University introduced Koopman Expectation-based optimization for spacecraft rendezvous guidance under parametric uncertainty. It is, in essence, a structured linear world model in lifted latent space, closely related in spirit to RSSM architectures in the Dreamer lineage.
I enjoy building intelligent systems that are both innovative and practical.
Generative AI systems, agentic architectures, and the research pipelines behind them.
Autonomous multi-agent workflows performing complex reasoning, planning, and dynamic tool use. Self-correcting agents orchestrating APIs, databases, and code execution through MCP standardization.
Production retrieval pipelines with semantic splitting, hybrid chunking, and vector database tuning. Agentic and corrective RAG variants for dynamic retrieval and reasoning.
PEFT methods and RLHF for adapting models to domain tasks. Quantized, async inference pipelines with KV-cache reuse and GPU acceleration for low-latency serving.
Koopman Expectation methods for propagating uncertainty through nonlinear dynamics. Structured latent world models for guidance, closely related to RSSM architectures.
Turns research papers into verified, executable code. A LangGraph pipeline with Planner, Coder, and Critic agents, hybrid ChromaDB plus BM25 plus RRF retrieval, and sandboxed execution. RL fine-tuning is the planned next phase.
An AI assistant that streamlined research knowledge for the Spacecraft Robotics Laboratory using a RAG pipeline built on Gemini with structured document parsing.
A Streamlit app helping Canadian students estimate living expenses across cities, powered by Gemini Pro for contextual cost breakdowns.
Built during a Mitacs internship. Used NLP to extract and analyze adverse drug reactions from social data, processing over 30GB and improving ADR prediction accuracy by 15 percent.
Reinforcement learning based trajectory optimization under uncertainty, implementing actor-critic methods for spacecraft rendezvous guidance.
End-to-end project predicting domestic flight prices in India using Random Forest and XGBoost regressors, deployed as a Flask web application on Render.
Adaptive control framework for a 2-DOF robotic arm operating under time-varying payload uncertainties.
Classification models identifying key employee attrition factors and predicting churn risk, with actionable insights for retention strategy.
Two running series where I work through foundational AI books chapter by chapter and write up what actually clicked.
A beginner-friendly breakdown of the key takeaways from Chapter 1, and why building an LLM from scratch teaches you things no API ever will.
Read on Medium →How text becomes tokens, why the choice of tokenizer shapes everything downstream, and what byte-pair encoding is actually doing.
Read on Medium →Walking through self-attention from first principles, from query-key-value intuition to the multi-head implementation.
Read on Medium →Assembling the transformer blocks into a working architecture. Coming soon.
Why the shift from training models to building on top of them changes the whole engineering discipline, and what skills carry over from traditional ML.
Read on Substack →Training data, modeling architecture, post-training, and sampling. The four levers that determine what a foundation model can and cannot do.
Read on Substack →Why evaluating open-ended generation is genuinely hard, and the methods that hold up. Close to my day job in frontier model evaluation.
Read on Substack →Moving from model evaluation to system evaluation. Publishing next week.
Thesis: Koopman Expectation-Based Optimization for Offline Spacecraft Rendezvous Guidance Under Parametric Uncertainties
View thesis →Thesis: Reinforcement learning framework for spacecraft low-thrust orbit raising
View thesis →Foundation in aerospace systems, propulsion, and flight dynamics.
IMS Proschool
Open to roles, research collaborations, and speaking engagements across AI, ML, and aerospace.