Ritu KanchiSoftware Engineer · Machine Learning

Distributed, event-driven ML system using Kafka microservices to generate real-time sports commentary with sub-600ms end-to-end latency. Integrated NLP, emotion modeling, and vector-based retrieval for 88% human-aligned commentary accuracy.

Cloud-deployable RAG system for large-scale legal queries, achieving 98% faithfulness and 87% answer correctness. Engineered semantic search, reranking, and caching over 3000+ documents, reducing response latency by 40%.

PDF Extractor! Extracts all the words from a pdf and renders it (also has highlight tagging) built with Nextjs, Render, Flask backend.

A complex dashboard for visualizing and interacting with large datasets, featuring server-side rendering and advanced filtering capabilities.

Advanced recommendation engine with a fine-tuned BART-Base model mapping unstructured descriptions to a 27-tag taxonomy, achieving 60% improvement in tag accuracy and 0.53 Precision@15 over T5 baselines. Optimized on consumer hardware using bfloat16 mixed-precision with <4% hallucination rate. Coming soon!

Inspired by my internship, a supply chain workflow optimization using AI ML( fine-tuned mistral 7B on ERP/CRM data).