I'm an AI & Automation Engineer based in Bengaluru, currently at SAP Concur where I design and own production LLM + RPA systems that handle thousands of enterprise cases every week. I care about one thing: automation that actually works at scale, not demos.

My stack lives in the LangChain/LangGraph/RAG ecosystem, deployed on SAP BTP with CI/CD through Honeycomb. Before this, I built across Python, TypeScript, and Go: desktop GUI apps, multi-agent SQL interfaces, full-stack web platforms.

These run in prod. Not side projects.

LLM + IRPA Triage Pipeline Parses case subject, description, and business rules to route 1,000+ cases/week to the right consultant queue automatically, replacing a fully manual process.

E2E Concur Implementation Agent Multi-agent pipeline: an analysis agent reasons through workbook configs and dependency ordering, a planning agent generates a JSON task plan, a critic agent flags gaps, and a chat agent handles human-in-the-loop review before a Playwright-based execution layer configures SAP Concur entities end-to-end.

Triage Validation Engine LLM comparison of bot-assigned vs consultant-modified inventory groups. Flags triage errors automatically, replacing 50-60 manual reviews per week.

Risk Analysis System (Pandas + Salesforce signals) Scores 3,000+ enterprise accounts across Safe / Should Monitor / Critical tiers using a rule-based weighted engine built on case spikes, escalations, and SLA signals.

Python · LangGraph · ChromaDB · MedEmbed · PySide6 · Gemini

Desktop app with a hybrid RAG pipeline: local ChromaDB vector store (medical-specialized MedEmbed embeddings) + live Google search, fed into a LangGraph ReAct agent. Every response follows a strict safety template: Triage, Condition, Steps, Medicines, Citations. Severe symptoms trigger emergency escalation. QThread workers keep the GUI fully responsive during ingestion.

Python · LangChain · ChromaDB · PySide6 · PostgreSQL · MySQL · SQLite · Oracle · MSSQL

Multi-agent system that turns plain English into SQL across five database types. A question enhancer agent refines the query and suggests visualizations; a SQL agent generates and self-corrects queries (up to 5 retry iterations); a response agent builds the final answer; a visualization agent generates Plotly charts. Results download as .xlsx. Document uploads enrich ChromaDB context. Credentials encrypted at rest.

Next.js 14 · TypeScript · Convex · Clerk · Google Generative AI · Tailwind CSS · Recharts

Live at vidharith.vercel.app. Educators upload teaching materials; the platform auto-generates quiz questions via Gemini, runs real-time analytics as students respond, and surfaces per-student and class-wide performance breakdowns. Built on Convex for real-time data sync, Clerk for auth, and Shadcn/Radix UI for components.

Desktop art portfolio manager for artists. Built with PySide6. Watch the demo on YouTube

AI / ML

Backend & Data

Frontend & Full-Stack

Infrastructure & Tools