At Shyam Technologies, I worked on AIVOA, a CRM assistant whose API response time I reduced from 8 seconds to under 2 seconds using model quantization, GPU acceleration, and response caching. I also built its Generative AI pipeline, reducing manual documentation effort by 60–70%.
To turn unstructured CRM text into structured records, I implemented named entity recognition and transformer-based summarization with spaCy, PySpark, and MongoDB. I built a FastAPI backend with Redis caching and request batching for 500+ concurrent users, then deployed Docker containers on Kubernetes on AWS.
At Middleware Talents Ltd, I built 10+ Python ML and data-processing pipelines, reducing manual effort by 40% and improving data accuracy by 30%. I also built a billing platform that screens 500+ monthly transactions with ML-based anomaly detection at 95% accuracy.
At Packet Prep, I built Java and SQL billing systems with data-validation pipelines, cutting errors by 95% across 10,000+ daily transactions. In my Library Book Finder project, I built a LangGraph RAG workflow that handles 500+ daily queries across 10K+ documents, with sub-3-second latency and 99.5% uptime.

