Skip to content

Aravind Baranitharan, AI Engineer

Aravind Baranitharan, looking over his shoulder

Hello, I'm Aravind Baranitharan, an AI engineer building intelligent systems and impactful products.

Turning ideas into practical AI experiences.

(Scroll down)

AI Engineer · TCSAvailable · Nov 2026

i build rag pipelines

I build production LLM systems, agent workflows and the evals that keep them honest, then the product around them.

Start a project View the workDownload resume
  • TCS · RAG test-case generator

    less manual test-writing

    Test scenarios generated from enterprise documents by a local Mistral RAG service.

  • TCS · Figma-to-code agent

    faster screen development

    50–70% shorter build cycles with a multi-agent LangGraph code generator.

  • Smart India Hackathon 2026

    AI prototypes evaluated

    Internal evaluator for the AI domain, judging ideas and working prototypes.

  • LangGraph
  • LangChain
  • FastAPI
  • FastMCP
  • Fine-tuning
  • Quantization
  • RLHF
  • GNNs
  • Pinecone
  • Neo4j
  • Ragas
  • Mistral
  • Gemini
  • Hugging Face
  • Next.js
  • Docker
  • AWS

02 / About

who builds this.

I build production LLM systems, retrieval pipelines and the evals that keep them honest.

I'm an AI engineer at Tata Consultancy Services (ID&A unit) in Chennai. By day I ship RAG microservices, multi-agent code generators, knowledge-graph experiments and graph-neural-network forecasting for enterprise teams.

Before TCS I was founder-in-residence at Zemuria, leading Kix News, an AI platform that collates sources and checks news for bias. I also mentor teams at TCS AI Fridays, helping them get from idea to working prototype fast.

what I do.

AI engineer building production GenAI: retrieval pipelines, agent systems, fine-tuned and quantized models, and the evals that keep them honest. Full-stack enough to ship the product around the model.

01 / 05

AI + LLM integrations

Retrieval, re-ranking and evals wired into your product, with latency and cost you can see.

  • RAG
  • Re-ranking
  • Pinecone
  • Ragas
  • Grounded answers from your own data
  • Quality tracked with evaluation sets
  • Clear cost and latency numbers
02 / 05

Agent systems

Multi-agent workflows with LangGraph, A2A and MCP servers that plug into Cursor and Copilot.

  • LangGraph
  • A2A
  • FastMCP
  • Guardrails
  • Agents that coordinate on real tasks
  • Tools exposed to the assistants your team already uses
  • Guardrails for security and code standards
03 / 05

Custom web apps

Next.js products with the AI layer threaded through, from first sketch to deployed and instrumented.

  • Next.js
  • FastAPI
  • TypeScript
  • Tailwind
  • Fast, accessible interfaces
  • Typed APIs and validated data
  • Launch with analytics in place
04 / 05

SaaS product build

From MVP scope to role-based dashboards and billing, the way Mark Bot runs content agencies today.

  • Auth
  • Dashboards
  • Billing
  • Deployment
  • A scoped MVP that proves value
  • Role-based access and billing flows
  • A roadmap driven by real usage
05 / 05

Model fine-tuning

Open models adapted to your domain with fine-tuning and RLHF, then quantized to run on your own hardware, plus graph neural networks where the data is a network.

  • Fine-tuning
  • RLHF
  • Quantization
  • GNNs
  • A model that speaks your domain's language
  • Smaller, cheaper models that run on-prem
  • Graph models for networks like rail and logistics

how I work.

  1. Week 1

    Scope

    Define the wedge. Cut anything that does not move a metric or a feeling.

  2. Week 2

    Build

    Production code from day one. UI, API and the AI layer threaded together.

  3. Week 3

    Ship

    Deployed, instrumented, handed over with docs and a roadmap for what comes next.

04 / Work

selected work.

Agent engineering · 2026

Figma-to-code agent

A multi-agent system that turns Figma designs into production-ready UI code.

ResultCut screen development lifecycles by 50–70%.

  • LangGraph
  • GPT-5.2-codex
  • A2A
  • FastMCP
  • Context engineering
50–70%
Faster screen development
A2A + MCP
Agents usable from Cursor and Copilot
Read the case study: Figma-to-code agent
LLM systems · 2025

RAG test-case generator

A RAG microservice that writes test scenarios from unstructured enterprise documents.

ResultReduced manual test-scenario creation effort by over 60%.

  • FastAPI
  • LangChain
  • Mistral (quantized)
  • Pinecone
  • Ragas
60%+
Less manual test-writing effort
On-prem
Model hosted locally for compliance
Read the case study: RAG test-case generator
Founder · full stack · 2025

Mark Bot

An operating system for content agencies: production, operations and billing in one role-based dashboard.

ResultWeekly operations meetings dropped from 90 to 15 minutes, and writers spend far less time scripting.

  • LLMs
  • Prompt engineering
  • Next.js
  • Full-stack SaaS
90→15
Minutes of weekly ops meetings
8 stages
Topic-to-publish pipeline
Read the case study: Mark Bot

More projects (8)

All projects

11

05 / Experience

experience.

  1. 01AI EngineerTata Consultancy Services (ID&A Unit)Aug 2025 – Present · Chennai · On-site
    • RAG test-case generation microservice: FastAPI, LangChain, quantized local Mistral, Pinecone, re-ranking and Ragas monitoring. Over 60% less manual test-scenario effort.
    • Figma-to-code multi-agent system on LangGraph with a custom layout parser, guardrails and context engineering. 50–70% faster screen development.
    • A2A communication layer and a remote MCP server (FastMCP) exposing code generation to Cursor and GitHub Copilot.
    • Knowledge-augmented generation PoC: SQL database → knowledge graph → semantic model → graph RAG.
    • Train-delay forecasting PoC for an Indian Railways (IRCTC) client pilot: SQL triggers feed an AI agent that turns live operational records into graph inputs for a Graph Neural Network inference engine.
    • AI Fridays mentor, helping teams prototype AI solutions quickly.
  2. 02Founder-in-ResidenceZemuria (Kix News)Jan 2024 – Sep 2024 · Remote · San Francisco
    • Led product development of Kix News, an AI platform for unbiased news through source collation and automated bias-checking.
    • Built the aggregation engine with Python, Gemini, Jina AI and Selenium across sources and RSS feeds.
    • Took the product through its investment phase.
  3. 03B.Tech, Information TechnologySri Manakula Vinayagar Engineering College2021 – 2025 · Puducherry
    • CGPA 7.29.

ML

  • Machine learning
  • Deep learning
  • CNNs
  • Graph Neural Networks (GNNs)
  • Computer vision
  • ML algorithms

GenAI

  • LLMs
  • RAG
  • Multimodal RAG & GraphRAG
  • KAG
  • Knowledge graphs
  • Fine-tuning
  • Quantization
  • RLHF
  • Re-ranking
  • Agentic AI
  • Multi-agent systems
  • A2A
  • MCP
  • Context engineering
  • Guardrails
  • NLP
  • Prompt engineering

Frameworks

  • FastAPI
  • LangChain
  • LangGraph
  • FastMCP
  • Flask
  • Next.js
  • Tailwind CSS
  • Bootstrap

Models

  • GPT-4 / Codex
  • Gemini
  • Mistral

Data

  • Pinecone
  • Neo4j
  • Firebase
  • REST APIs

Tools

  • Hugging Face
  • Ragas
  • Cursor
  • Claude Code
  • GitHub Copilot
  • Jina AI
  • Selenium

Cloud

  • AWS
  • Microsoft Entra ID (RBAC)
  • Docker
  • Git

Languages

  • Python
  • JavaScript
  • TypeScript
  • C
  • HTML
  • CSS

Foundations

  • Data structures
  • Distributed AI architectures
  • IoT

recognition.

  • 1st place

    Hackverse Hackathon

    PNB MetLife, India

  • Mentor

    TCS AI Fridays

    Certificate of Appreciation · Chennai, Nov 2025

  • Jury

    SDG AI Hackathon

    ETH Zürich × NTU Singapore

  • Evaluator

    Smart India Hackathon 2026

    AI domain · 40+ prototypes

notes.

  1. “Ship the loop, not the prompt.”

    On building

  2. “Models hallucinate. Pipelines don't have to.”

    On engineering

  3. “Vibes scale to zero. Evals scale to production.”

    On rigor