AI / ML Engineer

Sevesh SS

I build and ship end-to-end AI systems: medical image segmentation, local-first computer vision, and retrieval-augmented apps that run on your own machine.

B.Tech AI & ML · Kongu Engineering College · Tirupur, Tamil Nadu

0.86Dice score on Kvasir-SEG polyp segmentation
3rdPlace, NEXORA’26 Hackathon
1 moTo deliver CAMS, requirements to production

About

Training models, then shipping them.

I’m a B.Tech AI & ML student who likes the whole path: preparing data, validating a model, and putting it somewhere people can use it. I trained and deployed a medical image segmentation model to HuggingFace Spaces, built local-first computer vision and RAG tools, and delivered a full-stack academic management system in a month as an AI Intern.

I work in PyTorch, React and Python, and I care about keeping data private, so most of what I build runs locally.

Studying
B.Tech, Artificial Intelligence & Machine Learning, 2024 – 2028
College
Kongu Engineering College
Languages
English, Tamil, Hindi
Based in
Tirupur, Tamil Nadu, India

Experience

Where I’ve worked.

AI Intern

CubeAISolutions Tech Pvt Ltd
JAN 2026 – FEB 2026
  • Developed core AI-driven modules for the College AI Management System (CAMS), streamlining data preparation and record-management workflows.
  • Led full-stack development in React and Python, integrating every layer from the data layer to the UI.
  • Designed the MongoDB persistence layer and pipeline structure to speed up storage and retrieval of academic records.
  • Delivered CAMS end to end, from requirements to production deployment, within a one-month timeline.

Projects

Selected work.

Medical image AI

Polyp Segmentation

A U-Net with a ResNet34 encoder trained on Kvasir-SEG to find polyps, reaching 0.86 Dice and 0.76 IoU. A pretrained encoder and stronger augmentation lifted Dice from 0.77, and the model is deployed on HuggingFace Spaces for real-time inference.

PyTorchU-NetResNet34HuggingFace Spaces
View on GitHub →
Computer vision · RAG

ShuttleSense

Local-first badminton video analysis. YOLO11-pose, MediaPipe, TrackNetV3 and SegFormer power rally replay, shuttle tracking, body-pose overlays and heatmaps. A RAG coach chat (Qwen3 on Ollama, Qdrant) writes evidence-grounded match reports as PDF or HTML. Video and inference never leave your machine.

PyTorchOpenCVNext.jsTypeScriptOllamaQdrant
View on GitHub →
Knowledge vault

ReelToReal

A local pipeline that turns saved Instagram Reels and YouTube Shorts into timestamped, Obsidian-compatible notes using yt-dlp, faster-whisper and a vision-language model. Hybrid search combines Qdrant vectors with SQLite BM25 through Reciprocal Rank Fusion, with a RAG chat and a React dashboard on top.

FastAPIReactWhisperQdrantSQLite FTS5Ollama
View on GitHub →
Voice · Local AI

Local Voice

A local voice companion that understands English, Hindi and Hinglish. Speech goes in through Whisper, a Qwen model handles the conversation, and Kokoro speaks the reply, all behind a FastAPI backend and a React front end.

PythonFastAPIReactWhisperQwenKokoro
View on GitHub →
More experiments

Skills

What I work with.

ML / Deep learning

PyTorchTensorFlowOpenCVCNNLSTMU-NetYOLOMediaPipeWhisperRAGLLMs (Ollama)

Languages

PythonTypeScriptJavaScriptC++JavaOOP

Web & backend

React.jsNext.jsFastAPINode.js / Express

Databases

SQLMongoDBSQLiteQdrant (vector DB)

Tools & deployment

GitGitHubFFmpegHuggingFace SpacesCross-validation

Awards & training

Hackathons and recognition.

3rd Place
NEXORA’26 HackathonSathyabama Institute of Science and Technology · 1–2 Sept 2026
2nd Prize
Eco Warrior Award, FIESTA A’26KPR Institute of Engineering and Technology · Feb 2026
Participant
HackITon’26, 24-hour national hackathonKPR Institute & IEEE · Oct 2026
Participant
International Hackathon 360°-3.0KPR Institute & ECLearnix EdTech · Feb 2026 · represented KEC in the FIESTA A’26 AI/ML Project Presentation
Training
In-Plant Training, Data AnalyticsEnsemble Nexgen, Erode

Contact

Let’s build something.

Open to internships, research collaborations and hackathon teams.

11sevesh@gmail.com