Bharat Khanna

Senior Associate Data Science @Publicis Sapient

New Delhi, DL, IN
MOBILE NUMBERS
+91 *********19

Signup · Get unlimited contacts

WORK HISTORY

Oct 2024 — Present

Senior Associate Data Science @Publicis Sapient

View department →

Gurugram, IN

LLM-to-SLM Migration – Real-time Hierarchical Intent Detection: Replaced a large language model with a customBERT-based Small Language Model for multi-intent classification on live call transcripts, significantly reducing inference latency and cost. Designed a 3-tier hierarchical classification architecture supporting 1,800+ fine-grained intent classes, enablingaccurate real-time intent detection across complex taxonomies at production scale. Built end-to-end offline evaluation pipelines with controlled experiments to validate model improvements, driving iterativeperformance gains; achieved model accuracy exceeding 90% and delivered $1M in cap-ex savings. Enterprise Knowledge Search – RAG Platform: Architected a production-grade Retrieval-AugmentedGeneration (RAG) system enabling context-aware search across large internal knowledge repositories for support andoperations teams. Designed Model Context Protocol (MCP) server integrations to expose enterprise knowledge sources — includingdocument stores and internal APIs — as standardized tool endpoints, enabling agentic workflows to dynamically retrievecontext across heterogeneous systems without custom connectors. Implemented hybrid retrieval combining dense vector embeddings (FAISS) with BM25 keyword search, followed bycross-encoder re-ranking to maximize semantic relevance on complex multi-hop queries. Engineered scalable document ingestion pipelines with chunking strategies, metadata enrichment, and vector DB indexingsupporting heterogeneous enterprise document formats. Developed benchmarking frameworks measuring retrieval quality via Precision@k, Recall@k, and MRR; improvedPrecision@k from 75% to 90% and reduced hallucination rate by 40% through iterative optimization of the retrievalstack. Built agentic orchestration workflows using LangChain and Google ADK with multi-step reasoning, query decomposition and reduced average query response latency from 5s to 3.2s in production.

EDUCATION

N/A

Kulachi hansraj model school

XII, Science

2011 — 2015

Maharaja Surajmal Institute Of Technology

Bachelor's degree, Electrical, Electronics and Communications Engineering

ABOUT BHARAT KHANNA

Lead Generative AI & LLM Engineer with 9+ years of experience building large-scale AI systems across telecom, healthcare, and energy domains.I specialize in:→ RAG Architectures — hybrid retrieval, cross-encoder re-ranking, hallucination reduction→ Agentic AI — LangGraph, Google ADK, multi-step reasoning, dynamic tool selection→ Model Context Protocol (MCP) — enterprise knowledge source integrations→ BERT / SLM — hierarchical intent classification, fine-tuning, multi-label NLP→ LLM Engineering — prompt engineering, Azure OpenAI, semantic searchTech stack: · LangChain · LangGraph · LlamaIndex · AutoGen · CrewAI · MCP · BERT · HuggingFace · GCP Vertex AI· PythonCertifications achieved-> Azure Data Scientist (DP-100)-> Azure Fundamentals (AZ-900)-> Azure Data Fundamentals (DP-900)-> Azure AI Fundamentals (AI-900)-> Azure AI Engineer Associate (AI-102)

This profile is compiled from publicly available professional sources. Unifers is not affiliated with or endorsed by LinkedIn. Request removal of this profile.

Bharat Khanna — Senior Associate Data Science at Publicis Sapient in New Delhi, DL, IN | Unifers