Hi, I'm Srinivasan 👋
Senior AI Engineer specializing in Agentic AI — building multi-agent LLM systems, agentic RAG pipelines, and production GenAI infrastructure for regulatory and fintech domains.
SR

About

Senior AI Engineer with 3+ years of experience designing Agentic AI systems in production — multi-agent orchestration with LangGraph, confidence-gated self-reflection, tool-using LLM workflows, and Pydantic-enforced structured outputs. I architect agentic RAG pipelines with hybrid retrieval (dense + sparse + cross-encoder reranking), and instrument every node with Langfuse for forensic-grade observability over latency, token cost, and retrieval quality. At XYMA Analytics (IIT Madras), I shipped the Autonomous Financial Dispute Decision Engine — an agentic system that reasons over NPCI/RBI regulatory corpora to classify disputes and decide refund eligibility — plus the UPI Regulatory Dispute Assistant for grounded UPI dispute resolution. I obsess over retrieval precision, inference latency, token cost efficiency, and audit traceability.

Work Experience

X

XYMA Analytics, IIT Madras

April 2025 - Present
Senior Software Engineer – AI Systems
Architected production RAG systems over NPCI/RBI regulatory and industrial corpora using hybrid BGE-Small dense + SPLADE sparse retrieval with cross-encoder reranking in Qdrant. Designed the Autonomous Financial Dispute Decision Engine — an agentic AI system for refund eligibility and dispute classification using LangGraph state machines, confidence-gated self-reflection, and Pydantic-enforced JSON outputs via Groq/Llama 3.3 70B. Built the UPI Regulatory Dispute Assistant (RAG over NPCI/RBI). Instrumented full LLMOps observability with Langfuse (per-node latency, token cost, retrieval quality) and async audit logging to MongoDB Atlas. Deployed auto-scaling inference on AWS (Lambda, API Gateway, S3, ECR) and GCP (Cloud Run) via Docker + GitHub Actions. Optimized context budgeting and Redis semantic caching to cut redundant LLM calls and per-request token cost.
X

XYMA Analytics, IIT Madras

March 2022 - March 2025
Junior Software Engineer – AI & Full-Stack
Bootstrapped the team's production RAG foundation — prototyped LLM integration workflows, benchmarked OpenAI APIs, prompt chaining, and embedding-based similarity search. Built vector embedding workflows with sentence-transformer models and ChromaDB for internal semantic search across enterprise documents. Delivered full-stack applications (Python, Node.js, React.js, TypeScript) across industrial, analytics, and HR domains with REST APIs and MongoDB/PostgreSQL schemas. Designed high-frequency IoT sensor ingestion pipelines and real-time monitoring dashboards for manufacturing clients. Built modular Python microservices that became the architectural foundation for the production RAG systems shipped in the senior role.

Skills

Agentic AI
Multi-Agent Orchestration
LangGraph
LangChain
Tool-Using LLMs
Self-Reflective Agents
RAG
Agentic RAG
Hybrid Search (Dense + Sparse)
Cross-Encoder Reranking
Pydantic Structured Outputs
Prompt Engineering
LLM Guardrails
Semantic Caching
Python
OpenAI API
Groq (Llama 3.3 70B)
BGE-Small
SPLADE
FastEmbed
Qdrant
ChromaDB
Pinecone
Weaviate
FastAPI
AsyncIO
Crawl4AI
Langfuse
LangSmith
AWS (Lambda, SageMaker, S3, API Gateway, ECR)
GCP (Cloud Run)
Docker
Kubernetes
GitHub Actions
MongoDB Atlas
Redis
PostgreSQL
React.js
TypeScript
My Projects

Agentic AI systems I've shipped

Multi-agent LLM workflows, agentic RAG pipelines, and production LLMOps infrastructure for regulatory and fintech domains.

Autonomous Financial Dispute Decision Engine

Autonomous Financial Dispute Decision Engine

Agentic AI system for automated refund eligibility and dispute classification over RBI/NPCI regulatory corpora. Built with LangGraph state machines, confidence-gated self-reflection, and Pydantic-enforced structured JSON outputs via Groq/Llama 3.3 70B. Grounded in hybrid retrieval (BGE-Small + SPLADE) with cross-encoder reranking for forensic-grade auditability.

LangGraph
Groq / Llama 3.3 70B
Pydantic
Qdrant
BGE-Small
SPLADE
FastAPI
Langfuse
MongoDB Atlas
UPI Regulatory Dispute Assistant

UPI Regulatory Dispute Assistant

RAG-based contextual querying system over NPCI/RBI regulatory datasets for UPI dispute resolution. Built async PDF ingestion pipelines, semantic chunking, vector embeddings, and grounded LLM response generation with retrieval quality scoring and per-request token cost tracking.

RAG
LangChain
ChromaDB
Qdrant
FastAPI
OpenAI API
Redis
Langfuse
Rewardive

Rewardive

Mobile app I'm building that maximizes credit card rewards by recommending the best card to use for every purchase. Aggregates offers from 50+ Indian banks across 10,000+ active deals and 500+ spending categories — powered by QR scanning, merchant category classification, and AI-driven offer matching.

Flutter
Dart
AI Offer Matching
Merchant Category Classification
QR Scanning
FastAPI
Python
Namma Wallet (Open Source)

Namma Wallet (Open Source)

Open-source Flutter wallet for Indian users — an alternative to Apple/Google Wallet that stores digital tickets and passes (TNSTC buses, IRCTC trains, event passes) parsed from SMS, PDFs, QR codes, and clipboard. Supports the .pkpass standard. Contributing as part of the namma-Flutter community.

Flutter
Dart
Syncfusion PDF
pdfrx
Fastlane
FVM
Hackathons

I like building things

During my time in university, I attended 1+ hackathons. People from around the country would come together and build incredible things in 2-3 days. It was eye-opening to see the endless possibilities brought to life by a group of motivated and passionate individuals.

  • H

    Hacknight Hackathon

    Chennai, TamilNadu

    Developed a web application one card verfication for all your education and goverment certification using decenteralized serviced called crust to for encrytion for storage and web3 login using polygon.
Contact

Let's build with AI

Working on a RAG pipeline, an agentic workflow, or production LLMOps? Reach me at srinivasan.r830@gmail.com or DM me on LinkedIn. I'll ignore soliciting.