VIEW →

MADHAV
SUNIL

Software Engineer / AI/ML Engineer / MLOps Engineer / Published Researcher / Robotics Engineer

SCROLL ↓

About

Madhav Sunil is an AI engineer in Trivandrum who builds things that have to explain themselves: small language models for industrial machines, a wheelchair that climbs stairs, and a village of 36 people who gossip.

Right now

  • Building governed retrieval and small models for industrial machines at NeST Digital.
  • Auditing my old trading bot with Oracle Autopsy, after it claimed ₹10,000 became ₹1,60,000 on a flat market.
  • Keeping Rumourtown running: whisper a rumour and watch 36 villagers pass it on.

Open to AI engineering roles and research collaborations.

I'm a Software & ML Engineer with a B.Tech in AI & Robotics from VIT Chennai, currently building production AI systems at NeST Digital STC.
My work spans the full stack of intelligent systems: small language models that map industrial device data with measured precision and human review, document-to-data pipelines for engineering BOMs, self-hosted LLM platforms, and cloud-native Kubernetes infrastructure with full observability. I'm also a first-author researcher published in Springer Nature, bridging academic ML with real-world clinical impact.
I thrive at the intersection of AI research, DevOps, and robotics, building things that are both technically rigorous and genuinely useful.

Work

May 2025 – Present
Software Engineer
NeST Digital STC
  • Industrial IoT AI mapping: built the AI services that onboard industrial devices and standardise their telemetry, with a local 4B model reading OEM manuals and a 0.5B model mapping raw fields to a 10-industry canonical catalogue behind Designer/Challenger review, physical-unit checks and human approval. 100% precision on 90 held-out cases, 34 ms p99 latency, 986 automated tests, validated across 8 protocols (OPC-UA, MQTT, BACnet/IP, SNMP, REST, FTP, XMPP, OCPP).
  • RFQ BOM intelligence: designed a four-stage pipeline turning customer technical data packs (PDF, Word, Excel, drawings, STEP/IGES) into structured engineering BOMs, with a 194-field extraction contract (74/74 validation checks) over 6,210 raw observations. Benchmarked PostgreSQL against Neo4j: 36/36 queries matched an independent oracle and PostgreSQL ran 3 to 4.6x faster at 124,202 records.
  • Self-hosted AI platform: built an on-premise ML platform replacing a cloud AI accelerator suite (FastAPI, Next.js 14, PostgreSQL, Ollama, MinIO, Redis), with 201 API endpoints across RAG, NLP-to-SQL, fine-tuning, model registry, deployments and MCP, 17 Docker services with local, multi-tenant SaaS and edge modes, an agentic ML assistant (DeepSeek-Coder 33B) and 391 automated tests.
  • ContractRAG: designed governed retrieval that binds every answer to an approved knowledge contract. In a live poisoned-corpus benchmark with an LLM generator it held knowledge leakage at 0% (vector and hybrid search: 60%), out-of-scope retrieval at 0% (vs 18%) and contract compliance at 100% (vs 40%), with zero unsafe or unsupported answers.
  • Autonomous stair-climbing wheelchair: built the ROS2 Humble stack (10 packages, 61 nodes, 27 custom messages) for a hybrid wheelchair with mecanum drive and tri-star climb wheels: a 19-state controller using dual OAK-D depth cameras for stair-edge, drop-off and marker detection, IMU seat levelling, an STM32 motor bridge and an MQTT cloud layer. In a continuous 15.6-minute autonomous Gazebo mission it climbed 3 flights, descended 1, made 5 vision-guided turns and avoided an obstacle with zero manual takeovers.
  • NLP-controlled robotic arm: 6-DOF arm driven by text or voice, using Whisper large-v3 and Llama 3.3 70B to produce servo plans for custom Arduino firmware, plus YOLOv8 vision with homography calibration for automatic targeting.
  • Warehouse digital twin: headless FreeCAD-to-Unity pipeline converting CAD assemblies to glTF 2.0 and their mechanical joints into runtime Unity constraints.
  • Reverse-engineered legacy systems and delivered end-to-end automation pipelines using RPA and Workato.
  • Won First Prize in a company-wide Hackathon for an SQL-AI driven analytics platform.
Jun 2024 – Jul 2024
Summer Research Intern
Vellore Institute of Technology
  • Conducted a comprehensive study on the relationship between pulmonary embolism (PE) and heart failure.
  • Developed a hybrid deep reinforcement learning strategy (A3C + PPO) to enhance clinical model performance.
  • Applied SMOTE and LIME, work that led directly to a peer-reviewed publication in Springer Nature.
Apr 2023 – May 2025
Data Science Team Mentor
IEEE RAS, VIT Chennai
  • Mentored 20+ junior members in ML, deep learning and data pipeline design across hackathons and research projects.
Oct 2023 – Dec 2023
Student Intern
Tata Consultancy Services (TCS)
  • Delivered two end-to-end computer vision projects covering image classification and object detection pipelines.
Aug 2023 – Dec 2023
Student Intern
INVIS Pvt. Ltd.
  • Built an AI-powered speech-to-text module for automated meeting transcription using NLP pipelines.
  • Developed intelligent web scraping systems for structured data extraction and downstream analytics.

Tools I reach for

01
ML & AI
Deep LearningReinforcement LearningXGBoostTransformersTransfer LearningNLP & VisionLLaMAHugging FaceOllamavLLMQwenDeepSeek-CoderGroqClaude APIRAGMCPAI AgentsLLM EvaluationYOLOv8WhisperSMOTELIME
02
MLOPS & DEVOPS
Docker ContainerisationKubernetes OrchestrationHelm ConfigurationsTerraform IaaSCI/CD PipelinesModel DeploymentPrometheusGrafanaMLflowMicrosoft AzureCloudflare Workers
03
SWE & SYSTEMS
REST APIsSystems DesignData EngineeringFeature EngineeringWeb ScrapingRPA AutomationsFastAPIJinja2PydanticNext.jsTypeScriptC#PostgreSQLNeo4jWebSocketsSQLWorkato
04
ROBOTICS & INDUSTRIAL IoT
Simultaneous LocalisationEmbedded SystemsSensor FusionServo ControlUnity 3D SimulationROS2GazeboSTM32OPC-UAMQTTBACnetArduinoRaspberry Pi 5ESP8266FreeCAD

Research

FIRST AUTHOR · SPRINGER NATURE · DISCOVER ARTIFICIAL INTELLIGENCE · 2025
Integrating Advanced Deep Reinforcement Learning and Explainable AI for Enhanced Pulmonary Disease Detection

Developed an interpretable deep RL framework (PPO + A3C) for early detection of pulmonary-embolism-induced heart failure. Integrated SMOTE-based class balancing and LIME-driven explainability to deliver clinically transparent, reliable predictions. Peer-reviewed and indexed in Springer Nature.

DOI: 10.1007/S44163-025-00560-X →

Inside ContractRAG

Same model, same question. Only one of them is allowed to read everything.

Ten maintenance questions went to the same LLM, over documents with an off-topic file planted among them. Plain vector search reads whatever ranks highest. ContractRAG reads only what an approved knowledge contract allows, checks every claim against that evidence, and refuses when it cannot. These are the recorded runs.

Question 1 of 10

What PPE is required for standard compressor maintenance tasks?

Plain vector search

Reads whatever ranks in the top five.

no contract
  1. S1Safety SOP
  2. S2Safety SOP
  3. M1Maintenance manual
  4. M2Maintenance manual
  5. X1Off-topic policy fileoutside contract

Safety glasses and cut-resistant gloves are required for standard compressor maintenance tasks according to SOP-MNT-014. Steel-toe boots and hearing protection may also be needed depending on the task's specific hazards as mentioned in Source 1: compressor_motor_safety_sop.md.

Reviewer: correct and grounded21.2 s

This answer was written with 1 off-topic passage in its context.

ContractRAG

May read the safety SOP and the maintenance manual. Nothing else.

X1kept out by the contract
  1. S1Safety SOP
  2. S2Safety SOP
  3. M1Maintenance manual
  4. M2Maintenance manual
  5. M3Maintenance manual

Safety glasses [Source 1] Cut-resistant gloves [Source 1] Steel-toe boots [Source 1] Hearing protection [Source 1]

Reviewer: correct and grounded8.8 s
All ten questionsvector searchContractRAG
Answers written with off-topic text in context60%0%
Retrieved passages outside the contract18%0%
Answers that stayed inside the contract40%100%
Questions handled correctly100%50%
Median time to answer16.3 s7.7 s

The trade-off is real: ContractRAG refused four questions the documents could answer, so it handled half the set correctly where plain search handled all of it. In exchange it never read text it was not cleared for. Ten questions, eight passages, one recorded run, every answer labelled by a human reviewer. It shows the mechanism, not scale.

Things I've built

Rumourtown landing screen: a soft dusk sky with the title and an Enter village button
01
Rumourtown
A browser game where 36 villagers pass on a rumour you whisper, and their beliefs, friendships and grudges shift as it spreads. Runs on rumormind, my dependency-free TypeScript library for beliefs with provenance, using research-backed rules for trust, repetition and corrections. Deterministic and replayable; optional AI voices conversations behind a fail-closed spending cap.
rumourtown.dev ↗
2026 · Multi-Agent Simulation
→
The Aarangi shop home page with handwoven sarees and a Shop the collection button
02
Aarangi
Online shop and inventory system for Aarangi, a handwoven saree boutique in Kollam, Kerala. Cloudflare Workers with Razorpay checkout (UPI, cards, EMI), 15-minute stock holds, sale pricing and a Google Shopping feed. A private staff portal with Access sign-in, D1 inventory and R2 photos gives every piece a permanent code.
aarangi.com ↗
2026 · E-commerce / Full Stack
→
Oracle Autopsy chart: the old bot reports ₹10,000 climbing to ₹1,60,000 while the reference ledger stays flat
03
Oracle Autopsy
My old trading bot reported ₹10,000 becoming ₹1,60,000 on a market that never moved. This is the tool I built to catch that. A strategy only gets a GO from it after surviving an independent ledger, look-ahead checks, real Indian costs (STT, GST, stamp duty), a sealed final year and 1,000 random-entry traders. It also sends paper buy and sell alerts for NSE stocks during market hours. Python, 333 tests, never places orders.
github.com/Maddy824/oracle-autopsy ↗
2026 · Quant Research / Python
→
04
AI Infrastructure Generator
Points at any repository and automatically plans + generates production-ready Docker, Compose, Kubernetes, Helm, Terraform, and CI/CD configs using AI. Strict three-phase architecture: Understand → Plan → Generate. Full observability with Prometheus & Grafana.
2025 · MLOps / Automation
→
05
DL-In-Medic
Deep learning system for real-time dermatological screening using phone camera input with 3D augmentation and medical imaging classification.
2025 · Healthcare AI
→
06
INVI Social Scraper
Automated web scraping system targeting social media platforms for structured data extraction, enabling downstream analytics and intelligence workflows.
2025 · Data Intelligence
→
07
NLP Intelligence Agent
Autonomous agent that captures meeting audio, transcribes speech-to-text, generates AI-powered summaries, and auto-distributes updates to participants.
2025 · NLP / Automation
→

Education

2021 – 2025
B.Tech in Computer Science
VIT Chennai · AI & Robotics
Graduated with a CGPA of 8.46, specialising in AI theory, computational algorithms and applied robotics.
2019 – 2021
Class XII, CBSE
Suguna PIPS Coimbatore
2015 – 2019
Class X, CBSE
SN Trust Central School

Certificates

◆
Generative AI Engineering & Fine-Tuning Transformers
IBM • Mar 2026
◆
Machine Learning in Production
DeepLearning.AI • Mar 2026
◆
Google AI Essentials V1
Google • Dec 2025
◆
Problem Solving & Software Engineer
HackerRank • Aug 2025
◆
Enterprise Automation Certification
Workato • Jun 2023
◆
AI for Everyone
DeepLearning.AI • Feb 2023

Say hello

I'm open to exciting roles in AI engineering, MLOps, software engineering, or research collaborations. Whether you have a project, opportunity, or just want to talk tech, reach out.
Let'sbuildsomethingaccountable.
copied
madhavsunil99@gmail.com