At Krish TechnoLabs, I build and ship production ML and LLM systems for digital commerce, including a multi-tenant Bayesian dynamic pricing engine, a RAG-based conversational analytics assistant, and multimodal recommendation pipelines.
I improved production performance by removing DynamoDB reads through demand caching, optimizing recommender inference with ONNX and precomputed rankings, and reducing decision-engine latency with TTL caching, non-blocking logging, and connection-pool tuning. I also benchmarked Gemini and AWS Bedrock models for latency, cost, and hallucination rate while making analytics responses permission-aware.
Earlier, I built recommendation APIs, migrated marketing attribution workloads from BigQuery to AWS Athena and Glue, and developed ML applications for housing prediction and accessibility research. I enjoy taking ML systems from data ingestion and modeling through evaluation, deployment, and measurable operational improvements.
