Profile
GenAI Engineer Profile
I am Nitish Jha, a GenAI engineer and AI/ML Computational Science Senior Analyst at Accenture.
I build enterprise AI systems that combine agentic orchestration, retrieval, tool-based reasoning,
and backend engineering into production-ready products.
My recent work covers SDLC automation, code generation, knowledge retrieval assistants, and
LLM-powered APIs using LangGraph, CrewAI, AutoGen, Google ADK, FastAPI, Flask, and Azure-based delivery workflows.
Current Focus
- Agentic AI systems that break work into tool-driven multi-step flows
- RAG pipelines with stronger retrieval relevance and grounded generation
- LLM application backends that move from internal demos to production usage
Education
Vellore Institute Of Technology (VIT), Vellore, India
B.Tech in Computer Science and Engineering
2019 – 2023 | GPA: 8.63
Work
GenAI and Research Experience
Accenture
August 2023 - Present
AI/ML Computational Science Senior Analyst | Bengaluru, India
- Worked on enterprise GenAI products for SDLC automation, code generation, and knowledge retrieval assistants.
- Built and productionized Agentic Studio, an enterprise GenAI platform that dynamically generates relevant agents, assigns tools, and orchestrates them based on user-defined tasks. The studio enables users to select from frameworks like CrewAI, AutoGen (AG2), LangGraph, and Google ADK to automate multi-step planning and execution.
- Developed a multi-agent code generation platform using FastAPI + MCP (Model Context Protocol) + A2A (Agent-to-Agent) collaboration, reducing manual engineering effort by 60% and accelerating delivery by 35%, saving ~200+ engineering hours per quarter.
- Delivered Flask + LangChain LLM APIs (GPT-4, Gemini, Claude) to transform user stories/requirements into structured technical outputs, cut manual effort by 40% and reduced output inconsistencies by 30% via prompt engineering and LLM evaluation pipelines.
- Built a centralized LLM gateway across 3 AI providers with database-driven provider selection and secrets manager integration, cutting provider-switching overhead by ~60% and unifying credential management for 3+ teams.
- Designed and optimized production RAG pipelines (ingestion, chunking, embeddings, retrieval, grounded generation) using Chroma + PGVector, improving answer relevance by 25%, while owning POC to production delivery, CI/CD (Azure DevOps), testing (Postman), and mentoring 2–4 juniors to reduce rework by 20–25%.
Projects
Applied AI Projects
Hybrid RAG Document QA Agent
LangChain, BM25, FastAPI, ChromaDB
- Built a Hybrid RAG QA system over uploaded PDFs combining dense semantic retrieval (ChromaDB) with BM25 sparse retrieval, reranking, and score fusion for grounded, citation-backed answers.
- Tuned chunking and retrieval parameters (chunk_size, overlap, top-k, fusion weight alpha) to improve relevance; added configurable OpenAI model selection with per-request API key handling.
Autonomous Business Intelligence Agent
LangGraph, MCP, GraphQL, Python
- Built a stateful BI agent using LangGraph and Model Context Protocol (MCP) via GraphQL to dynamically extract, clean, and analyze cross-board CRM data based on 17 intent-classified natural language queries.
- Optimized LLM context window usage by 96% and reduced inference costs 10x by engineering a deterministic Python analytics engine, eliminating mathematical hallucinations on messy, real-world data.