ASHISH KUMAR

Architected and built an enterprise RAG platform and Model Context Protocol (MCP) server, enabling natural-language search over 2M+ document chunks for 500+ users.

Engineered a hybrid retrieval system fusing BM25 and 1024-dim vectors via reciprocal rank fusion (RRF) and LLM query planning, optimizing vector-index memory by 7x via on-disk FAISS quantization to avoid a ~17x costlier infrastructure upgrade.

Developed and deployed an LLM pipeline extracting structured financials from audited PDFs and reconciling them against MIS data, utilizing hybrid deterministic + LLM mapping with layered validation to cut mis-mappings and inference costs.

Built a production LLM pipeline extracting structured freight demand from 6K+ messages/day across 800+ groups, adding RCA and defaults layers to recover failed extractions and tuning prompts to save ~$50/month in inference costs.

Architected an event-driven AWS backend with SQS-chained Lambdas, Django REST APIs on PostgreSQL, and a React dashboard, while mentoring 3 junior engineers on AI integration and backend development.

Launched a CRM product with an auto-calling system, saving ₹3 Cr in third-party costs and increasing calls per sales executive by 5%.

Built a live counseling product connecting learners to counsellors instantly, driving a 4% subscription conversion rate against a 1% baseline.

Owned the user-communications microservice, including product and marketing messaging, with multiple queue pipelines to ensure efficient delivery.

Shipped critical features that directly powered over 4 lead-generating marketing campaigns, contributing to user acquisition and engagement.

Developed robust backend and client-side features for the enterprise data-science platform, enhancing functionality and user experience.

Refactored critical legacy modules and fixed numerous production bugs, significantly improving system stability and performance.