MLOps AWS Streaming
Kinesis, Flink, and Lambda, wired together with Terraform, for ML predictions that can't wait for a batch job.
Diffusion Models DPM-Solver++ Research
Why naive diffusion sampling is slow, and what a faster ODE solver actually buys you.
Agents LangGraph LLMs
What changes when an LLM app needs to reason across steps and remember context, not just answer one prompt.
Drug Discovery GPT-2 Molecular Modeling
Treating molecules as strings, and conditioning a language model on the properties you actually want.
RAG LLMs Retrieval
Why I combined BM25 and dense retrieval instead of picking one, in an end-to-end RAG web app.
LLMs GPT-4 Recommendation Systems
Building a product recommendation engine on GPT-4 instead of collaborative filtering.
Diffusion Models Deep Learning Research
Notes on diffusion models, from the core intuition to a real bidirectional MRI synthesis pipeline.