Headline

ML & GenAI Project Lead | Built Agentic AI Platforms · RAG · LLM Orchestration from Scratch | LiteLLM · pgvector · AWS Bedrock | Targeting AI Architect Roles

About

I design and build AI systems end-to-end — from agentic orchestration to production ML pipelines. At Persistent Systems, I've independently architected and shipped two GenAI products from scratch. → Built a multi-modal agentic platform: intent classification, DAG-based multi-source planning, and tool dispatch across 7 specialised tools — document Q&A (RAG + pgvector), Excel analysis (pandas REPL), image/video generation (AWS Bedrock), and multi-source workflows. LiteLLM proxy, MongoDB, pgvector embeddings, SSE streaming, Azure AD auth. → Designed a 5-stage AI call analytics pipeline: GPU-accelerated transcription, 8 parallel LLM prompts per call (sentiment, fraud, resolution, churn) with DynamoDB caching, PostgreSQL star schema, and a Plotly dashboard with PDF/Excel export. → Backend developer on an enterprise multi-tenant agent orchestration platform — tool registry, audit logging, and prompt lifecycle on AWS Bedrock and FastMCP. → Computer vision OCR pipeline digitising handwritten clinical documents for medical trial ingestion — 80% accuracy. Previously: 11 years at Ericsson (fraud detection 0.96 AUC, 3,500+ hours saved via automation) and TCS (15% cost reduction across Retail, Manufacturing, Telco).

Location

Bengaluru

AI stack

antigravity

claude code

cursor

github copilot

langchain

langgraph

Work experience

Project Lead at Persistent Systems, May 2025 - Present · 1 yr 4 mos, Bengaluru, Independently architected and built two GenAI products from scratch; contributing as backend developer on a third enterprise platform: • Multi-modal agentic platform (Architect & Builder): Designed the full agentic orchestration pipeline — intent classification, context-based routing, DAG multi-source planning, and tool dispatch. Built specialised tools for document Q&A (RAG + pgvector cosine search), Excel analysis (pandas REPL), image and video generation (AWS Bedrock), vision LLM chat, and multi-source agent workflows. SSE token-by-token streaming delivery. Azure AD auth with in-memory JWKS key cache. MongoDB for conversations and users; pgvector for document and image embeddings; LiteLLM as the LLM proxy. Core orchestration layers include identity manager, memory manager, and observability tracer. • AI call analytics pipeline (Architect & Builder): End-to-end 5-stage pipeline built from scratch. GPU-accelerated transcription service with optional PII redaction (Presidio). LLM processing runs 8 parallel prompts per call — extracting sentiment, fraud signals, resolution status, reason for call, and caller info — with DynamoDB prompt-response caching via LiteLLM. PostgreSQL star schema warehouse with asyncpg. Analytics dashboard (Plotly + MUI) with PDF and Excel export. Azure AD JWT auth with silent token refresh. • Enterprise AI agent platform (Backend Developer): Contributing developer on a multi-tenant AI agent orchestration platform. Working on tool registry services (REST/SOAP/MCP), immutable audit logging (DynamoDB + S3), and prompt lifecycle management — supporting SOC2/GDPR/HIPAA compliance. Built on AWS Bedrock and FastMCP. • Clinical OCR pipeline: Computer vision pipeline digitising handwritten patient surveys for medical trial data ingestion using Tesseract — 80% accuracy, eliminating manual transcription.

Project Analyst at Tata Consultancy Services, Jan 2023 - May 2025 · 2 yrs 5 mos, Use Python techniques to automate the daily activities including analysis and RCA. Use Tableau for data visualizations and helping strategic decisions using the same. Use SQL for ETL

Data Analyst at Ericsson, Sep 2020 - Dec 2022 · 2 yrs 4 mos, Bengaluru, Karnataka, India, Used machine learning and deep learning techniques to automate the daily activities including analysis and RCA. Developed segmentation models using K-means Clustering for exploring new user segments with AUC of 95%.Achieved a 10% cost reduction of yearly expenditure by implementing visualization tools to track the license issues across the network. Formed model for prediction and identification of worst cells with .96 AUC Score. Recommendation of RCA using Neural networks with saving of 400+ hours .

Senior Engineer at Ericsson, Apr 2014 - Dec 2020 · 6 yrs 9 mos, Bengaluru, Karnataka, India, Worked as data analyst in preparation of daily dashboards and using SQL queries for data extraction. Analyzed consumer data usage trends to identify growth areas for project expansion. Prepared daily, weekly and monthly KPI’s report on Tableau and Power BI with technical information of network and management information related to team utilization, project completion status etc. Created multivariate regression-based attribution models to achieve sales target . Successfully achieved a 10% cost reduction of yearly expenditure by implementing visualization tools and data analytics to track the license issues across the networks.

Engineer at Ericsson, Jul 2011 - Apr 2014 · 2 yrs 10 mos, Bengaluru, Karnataka, India, Summary :Data Analysis and Interpretation of data as per customer requirements Developed automation tools, increasing the efficiency by 20%. Devised and implemented an operational strategy to reduce project. Turnaround time between business units; turnaround time reduced by 20%.

Education

Master of Science - MS, MS Data Science, Liverpool John Moores University

PGDDS, Data Science, International Institute of Information Technology Bangalore