Satya Himavanth T
GenAI Developer | Agentic AI, Multi-Agent Systems, RAG & Document AI
Hyderabad
satyat2002@gmail.com
GenAI Developer at Tata Consultancy Services specializing in enterprise AI applications, agentic workflows, LLM orchestration, RAG, and Document AI. Experienced in building and optimizing solutions using LangGraph, LangChain, DSPy, Google Cloud Document AI, FastAPI, and vector databases. Delivered measurable engineering improvements, including reducing average LangGraph workflow execution time by 75% and improving SQL-agent accuracy from 86% to 97%. Built human-in-the-loop document processing pipelines, an AI assessment platform used by internal TCS teams, and prototypes for IT ticket automation and RAG evaluation. Interested in building reliable, scalable AI systems that turn LLM capabilities into useful business outcomes.
AI Engineer / Research Engineer – GenAI & Autonomous Agents
Tata Consultancy Services (TCS)
- Cisco — Optimized a custom LangGraph analytics workflow with approximately 6 agents and 20 graph nodes, reducing average execution time from 120 seconds to 30 seconds (75%) by eliminating unnecessary LLM calls and redundant MCP server/tool initialization.
- Cisco — Integrated the Jev decision model from TypeSafe for user-intent classification; implemented role-based data access and shareable KPI visualizations and summaries for collaborative business analysis.
- Zebra — Built a GCP Document AI pipeline for 26 document types, processing approximately 1,500 documents daily through threshold-driven classification, splitting, extraction, retraining, and human-in-the-loop review.
- Zebra — After retraining on approximately 5,000 documents, improved classifier accuracy from 70% to 100%, splitter accuracy from 86% to 100%, and extractor accuracy from 64.5% to 97.8%; deployed automatic Cloud Run processing triggered by configurable document-count thresholds.
- Infineon — Developed the SQL-agent component of a DSPy-based meeting intelligence solution, converting Microsoft Teams transcripts and meeting metadata into structured MySQL data and Qdrant-based semantic retrieval; improved SQL-agent accuracy from 86% to 97%.
- Infineon — Implemented SQL generation, DSPy prompt optimization, query validation, execution, result summarization, and evaluation using GPT-5 and GPT-4.1, supporting question-based routing to SQL, vector retrieval, or both.
- CoE — Designed and developed an AI assessment platform used internally by TCS, with approximately 5,000 candidates assessed; supports MCQ, text, coding and SQL assessments, LLM-based text evaluation, question categorization, and AI-assisted question generation.
- CoE — Built an IT operations agent prototype using historical ticket resolutions and internal documentation, achieving approximately 90% automatic ticket resolution without human assistance, as estimated for the prototype.
- CoE — Developed a RAGAS-based evaluation platform and LangChain connector for capturing evaluation data, comparing runs across users and projects, and helping developers identify areas for improvement.
- CoE — Built a ServiceNow agent enabling natural-language retrieval and updates of records, including ticket resolution, escalation, and reassignment; delivered additional PoCs involving Gmail agents, RAG, and agent evaluation.
- Deployed and managed AI systems on GCP using Cloud Run, Cloud SQL, VPCs, load balancers, Docker, and CI/CD workflows.
AI-Powered Assessment & Training Evaluation System
Built a full-stack AI assessment platform with automated MCQ, coding and SQL evaluation, LLM-based text-answer scoring, document-driven question generation, duplicate detection, batch assignments, and results analytics. Used by internal TCS teams for GenAI training assessments.
Built a framework-agnostic LLM observability platform that captures LLM calls, agent runs, tool invocations and multi-agent traces. Includes framework integrations, asynchronous trace collection, a FastAPI backend, a live dashboard, and project/user attribution for debugging and analysis.
Built an agent governance application for managing agent registration, user roles, team-based access, agent request approvals and audit logs. Includes an orchestrator that routes complex requests to authorized specialist agents.
Developed interoperable AI agents using the Agent-to-Agent protocol, with streaming, task lifecycle management, authentication, multimodal capabilities, push notifications, and a dashboard. Built a CLI test harness covering nine protocol and agent capability checks.
Implemented a local-first multimodal RAG pipeline that indexes text and images, retrieves relevant evidence from a shared vector index, and produces grounded answers. Supports local retrieval and extractive answering with optional hosted LLM generation.
B.Tech with Honors
R.V.R. & J.C. College of Engineering · 8.65
AWS Academy
Jun 2022Cloud and DevOps
EPAM
Jun 2023Java Full Stack
Wipro
Oct 2023
Attendee · Oct 24, 2026