Top Rated Freelancer · taking on new projects

SundarBalamurugan

Computer Vision & AI Engineer — Tiruppur, India

I build vision systems that ship: OCR and document intelligence, object detection, pose and gesture recognition, and the FastAPI / GPU pipelines that run them at scale. Independent, hands-on, and working directly with founders and teams across 22 countries through my studio, TheMadrasTechie.

See my work Start a project

70+

Clients

Countries

20k+

Students taught

1M+

Images / week in production

Add assets/sundar.jpg to show your photo here

Sundar BalamuruganFounder, TheMadrasTechie

Scroll

What I do

Vision systems, end to end

Object detection & tracking

Custom YOLO models trained on your data, from labelling through to a served endpoint. Retail shelves, PPE compliance, drones, security footage — anywhere a camera needs to count, find or flag something.

YOLOv8UltralyticsOpenCVTensorRT

OCR & document intelligence

Four years on financial documents at FRSLABS, and now vision-language models in production. Invoices, KYC, bills and forms turned into clean structured data — on a server or fully on-device.

Qwen-VLvLLMPaddleOCRML Kit

Pose, hands & gesture

Hand-landmark and body-pose pipelines for sign language, fitness, touchless interfaces and product demos. This is also where my published research sits.

MediaPipePose estimationASL

Vision + LLM applications

Products that pair a camera with a language model: logo search with CLIP and a vector database, shelf audits with vision LLMs, apps that read and reason about what they see.

CLIPQdrantClaude / Groq

Deployment & scale

FastAPI services, Docker, EC2 and GPU batch processing. I have built pipelines handling a million-plus images a week and cut batch runtimes 10× with GPU acceleration.

FastAPIDockerAWS EC2GPU batch

Cross-platform apps

When the model needs a front end, I build it — Flutter apps for iOS and Android that run vision on-device, plus dashboards in Plotly Dash for the ops team.

FlutterOn-device MLPlotly Dash

Selected work

Things I have built

Sign language · own product

Gesto — ASL fingerspelling recognition

Real-time fingerspelling from hand landmarks, packaged as a Python library on PyPI with a Flutter front end and a labelling tool for building new datasets. Grew out of my peer-reviewed ASL research.

MediaPipePythonFlutterPyPI

View on GitHub

OCR · vision-language model

Document OCR service, rebuilt on Qwen3-VL

Replaced a legacy TensorFlow / ONNX OCR service with a vision-language model served through vLLM, behind a FastAPI + Pydantic v2 API. Fewer moving parts, better extraction on messy documents.

Qwen3-VLvLLMFastAPIPydantic v2

Industrial safety · object detection

PPE compliance monitoring

YOLO models that watch factory camera feeds for helmets, vests and other gear, with a Plotly Dash dashboard for supervisors and a Flutter app for alerts on the floor.

YOLOPythonPlotly DashFlutter

Retail · vision LLM

Shelf analyser for retail audits

Photograph a shelf, get back gaps, facings and share-of-shelf. FastAPI backend calling vision models (Claude and Groq), deployed under the client's own brand.

FastAPIClaude visionGroq

iOS · similarity search

Logo scanning app

Point the phone at a logo and identify the brand. CLIP embeddings in a Qdrant vector store, a FastAPI backend, and Groq for fast text follow-ups — built for a client's App Store launch.

CLIPQdrantFastAPIiOS

Tooling · own product

YOLO Toolkit — annotate, train, serve

A Flutter annotation app paired with a FastAPI backend that trains YOLOv8 on your labels and serves the result, so a small team can go from raw images to a working detector without touching a terminal.

FlutterFastAPIYOLOv8

View on GitHub

Skanda AlgoCrypto trading bot for Delta Exchange India — Python, Docker, GitHub Actions, EC2, Telegram alerts.

Gesture-controlled web demosMediaPipe Hands in the browser: a virtual shelf navigator and a pizza builder for an FMCG pitch.

3D house generatorFastAPI + trimesh procedural geometry from a form, GLB export, viewable in the browser. Built for an architect.

Flutter apps

Apps I have shipped

Small, focused mobile apps — most of them offline-first, several with on-device vision.

🧾

Hotel bill OCRReads bills fully on-device with ML Kit and a Gemma 3 1B (int4) model. No server, nothing leaves the phone.Client app · on-device ML

🧩

Rubik's cube solverPoint the camera at each face; a vision model reads the colours and a 3D cube walks you through the solve.Vision + 3D rendering

✋

GestoASL fingerspelling recognition on the phone, built on the Gesto Python library and MediaPipe hand landmarks.Own product

📷

YOLO ToolkitAnnotate images on the go and send them to a FastAPI backend that trains and serves YOLOv8.Own product

🏫

ClassTrackMulti-role class and attendance management for a sports institute, with a super-admin panel.Client app

✅

AttendoOffline attendance tracker for small groups — local storage only, up to 100 people, no account needed.Offline-first

🔗

DailyDropSave links through the day, get one reminder to review them, comment, and mark seen or unseen.Offline-first

🌸

RecaloSave-and-recall link app with AI summaries and a cherry-blossom theme.Side project

🍛

MealMarkFood diary with a heatmap of your eating habits, place lookup via OpenStreetMap, and full export/import.Side project

About

One engineer, start to finish

Sundar Balamurugan

Computer Vision & AI Engineer · Founder, TheMadrasTechie

I started my first company at 17 and have been building computer vision systems ever since — four years on financial documents at FRSLABS, founding AI engineer and consultant for several startups, and 70+ freelance clients through Upwork where I hold Top Rated status. I have set up four engineering teams from scratch and prefer to stay hands-on with the code. Based in Tiruppur, Tamil Nadu.

LinkedIn GitHub Email

10+

Years building software

Engineering teams set up

10×

Batch speed-up via GPU

19k+

LinkedIn followers

Skills

Computer vision & deep learning

OpenCVYOLO / UltralyticsMediaPipeTensorFlow / KerasPyTorchObject detectionPose estimationOCRVision-language modelsCLIP

Backend & deployment

PythonFastAPIvLLMDockerAWS EC2GitHub ActionsTensorRTTritonQdrant

Apps & tooling

FlutterOn-device ML (ML Kit, Gemma)PyQt6Plotly DashClaude / Groq APIs

Experience

Jun 2026 – present

Contributor

gesto-ai

Open-source sign-language recognition — the Gesto library, labeller and Flutter app.

MediaPipePythonFlutter

May 2022 – Apr 2026

Computer Vision Engineer

FRSLABS

Four years of computer vision on financial documents: OCR, extraction and classification pipelines. Introduced GPU-accelerated batch processing that made large runs about 10× faster.

OCRGPU batchPython

Ongoing

Founder & Lead Engineer

TheMadrasTechie · Upwork Top Rated

Independent studio delivering CV and AI projects to 70+ clients in 22 countries. Founding AI engineer and consultant for several startups; built pipelines processing over a million requests and images a week; set up four engineering teams from scratch.

YOLOFastAPIVision LLMsFlutter

2015

Founder, at 17

Radssoon

My first company, started while still a student. Where I learned to ship.

Education & research

🎓

M.Tech, Artificial Intelligence — 2022

Amrita Vishwa Vidyapeetham

📄

Peer-reviewed publication

ASL recognition, published on ScienceDirect. Read the paper

⭐

Top Rated on Upwork

70+ clients across 22 countries.

📦

Open source on PyPI

Gesto and other packages, with Product Hunt launches.

Teaching

I also teach what I build

Four courses on Udemy, twenty thousand students

Computer vision and Python courses built from the same work I do for clients — practical, project-first, and kept current as the tools change.

4Courses

20,000+Students

CV · PythonFocus

See the courses

ConsultationsOne-to-one calls for teams choosing a CV approach, reviewing a pipeline, or unblocking a model that will not converge.

Team buildingI have hired and mentored four engineering teams from zero — happy to help you do the same.

Writing & talksRegular posts on computer vision and shipping AI products to 19k+ followers on LinkedIn.

Open sourceLibraries and tools released under TheMadrasTechie on GitHub and PyPI.