AI / LLM Engineer · Production multi-agent systems, RAG, and agentic workflows

I'm an AI/LLM engineer with 3+ years shipping production systems, currently building agentic AI at Planet Sustech. My work centers on multi-agent orchestration, typed tool-calling, and RAG — the unglamorous engineering that makes LLMs behave like reliable systems rather than demos.

I care about the parts that decide whether an agent survives contact with production: deterministic tool boundaries, server-derived analytics that kill hallucination, structured extraction from messy real-world inputs, and orchestration that degrades gracefully. Most of my recent systems run on LangGraph and AWS Bedrock, backed by NestJS/FastAPI services and Postgres/MongoDB.

Systems I've designed and shipped at Planet Sustech — ESG/climate domain, real users, real data.

A supervisor-routed agent (AWS Bedrock · LangGraph) exposing 21 typed tool-calling functions for natural-language ESG querying. Instead of NL-to-query generation, every answer flows through server-derived analytics, which eliminates LLM hallucination on numbers. Auto-generates executive dashboards with Excel/PDF export and RBAC across 4 roles. Cut manual analysis time by ~70%.

A vision + OCR pipeline (Bedrock) that extracts ESG metrics from 8+ document formats — PDF, Excel, scanned images, email — and auto-calculates Scope 1/2/3 emissions at 95% extraction confidence. Deployed on AWS ECS Fargate.

Ingests BRSR/XBRL filings for 987+ listed companies to produce SASB-weighted peer scoring, percentile rankings, gap-to-leader analysis, and 5 AI-generated improvement recommendations per company.

Scrapes supplier data from public sources, auto-builds custom assessments for data gaps, and runs an autonomous follow-up sub-agent to chase responses — targeting ~60% faster supplier onboarding.

A representative multi-agent pattern from my work — supervisor routing, specialized agents, a typed tool boundary, and analytics derived on the server rather than by the model.

flowchart TD
    U([Executive / User]) -->|natural language| R{Intent Router}
    R --> S[Supervisor Agent]

    S --> Q[Query Agent]
    S --> B[Benchmarking Agent]
    S --> I[Ingestion Agent]
    S --> D[Due-Diligence Agent]

    Q --> T[[Typed Tool-Calling Layer]]
    T --> DB[(ESG Data Store)]
    I --> V[Bedrock Vision + OCR]
    B --> X[BRSR / XBRL Filings]
    D --> W[Public Web Sources]

    Q --> O[Server-Derived Analytics]
    O --> RESP([Dashboards · Excel · PDF])

Principles I build by: typed tools over free-form generation · compute answers server-side, let the model orchestrate · sub-agents for autonomous follow-through · RBAC and auditability from day one.

Project	What it is	Stack
neverempty	Open-source Python eval harness that measures how often a tool-calling agent tells a user "no results" when a tool actually failed. Tools return Ok / Empty / Err instead of a bare list, faults are injected on purpose, and a CI gate fails the build when the misreport rate rises. Case study	Python · pydantic · pytest · GitHub Actions
NextRole	7-agent conversational job-search & career-coaching system with intent classification, multi-turn context management, and SSE streaming (90%+ routing accuracy)	FastAPI · Next.js · LangChain · OpenAI · Supabase
ESG Analytics Chatbot	RAG assistant converting natural language into MongoDB aggregation pipelines at 95%+ accuracy	Node.js · LangChain · MongoDB · RAG

Software Engineer — AI/LLM · Planet Sustech Private Limited · Aug 2024 – Present Built KarbonIQ and a suite of production ESG agents on AWS Bedrock + LangGraph. Owned agent architecture end to end — from tool-calling design and structured extraction to RBAC and deployment — and mentor a team of interns.

Associate Software Developer · Antino Labs Private Limited · Feb 2023 – Aug 2024 Engineered a Resource Management System for 400+ employees (Node.js, PostgreSQL, real-time analytics) and a social platform with AI-driven matching and Socket.IO/Agora realtime — 10k+ downloads in 4 months, +30% engagement.

Open to: AI/LLM Engineering · Agentic Systems · Full-Stack (AI focus)
📍 Gwalior, Madhya Pradesh, India  ·  🎓 B.Tech CSE, IPS College of Technology & Management (CGPA 8.2)