AI Engineer · Frontend Engineer · Dhaka, Bangladesh
I’m Saumik Saha Kabbya, a CSE graduate crafting LLM and computer vision systems, then shaping them into polished web experiences. I care about speed, clarity, and accessibility from prototype to launch.
Focus
LLMs & Vision
Flow
Ideas in, solutions out.
Signature
Fast, tactile, human-friendly UX.

Local-first vision assistance, LLM-powered recommendation systems, and multi-conversation chat platforms.
Syncing with GitHub…
Core stack
Full stack
Live
Real-time public GitHub activity and runtime signal from this deployed site.
Loading contribution map…
@Kabbya04Active repo
Loading...Last push
Loading...
Pushes (7d)
Loading...
Push streak
Loading...
Public repos
Loading...
Site RTT
Loading...
Syncing telemetry...
Experience
I work across backend intelligence and frontend polish, turning models into clear, responsive products.
AI Engineer & Frontend Engineer · Mar 2025 · Sept 2025
Developed computer vision and AI-powered client solutions with Python, FastAPI, and Groq AI; built YOLO/PyTorch detection pipelines and Next.js frontends for end-to-end AI features.
Education & Capstone
Balancing core CS fundamentals with hands-on AI systems built for real-world impact.
BSc in Computer Science & Engineering · Jan 2022 · Dec 2025
Focused on intelligent systems, full-stack engineering, and applied AI research. GPA: 3.66.
Jan 2025 · Present · Vision assistant for the visually impaired
Local-first vision assistant with ESP32-CAM wearable, YOLO26s and MediaPipe Active Guidance, BLIP/LLM scene narration, fall detection with guardian alerts, and a hands-free Web Speech API interface—87% object retrieval success rate.
Approach
Four defaults I follow on real projects.
I build the model and the UI. On AIris, Gravitas, and client work, splitting those two is usually where things fall apart.
Big work starts with a slice that holds up.
First move
Ask who uses it before I guess.
While building
Prototype before slides.
Done means
Accessible, or it stays open.
On a team
Updates that move work forward.
Projects
AI-driven tools, recommendation systems, and accessibility-first products that ship with clarity and speed.
Local-first vision assistant with ESP32-CAM wearable, YOLO26s and MediaPipe Active Guidance, BLIP/LLM scene narration, fall detection with guardian alerts, and a hands-free Web Speech API interface—87% object retrieval success rate.
A website for a Dhaka, Bangladesh-based small business selling gift items, accessories, and more.
AI-powered legal document understanding and grounded drafting with OCR, RAG, evidence-grounded draft generation, and iterative improvement from operator edits.
Website for the Bangladeshi cosmetics brand Hazel Studio, built with React, Vite, and Supabase.
Personalized recommendations for games, anime, TV series, and movies using Groq LLMs, TVDB for up-to-date metadata, and Supabase.
Document study tool that transforms documents into interactive guides with summaries, QA, chat, and TTS via Groq and ElevenLabs.
Multi-conversation AI chatbot platform with contextual sub-conversations and modular backend design.
Content-based game recommendation engine using Bag of Words, TF-IDF, and SVM to suggest similar video games with likelihood predictions.
Contact
Open to collaborations in AI, computer vision, and product-focused web builds. Share a brief and I’ll reply fast.
Email me
Click below and your mail app opens a draft addressed to me. Tell me what you’re building, what you need, or what caught your eye.
Direct line
saumik.kabbya@gmail.comVibe With The Dev
Put this on while you scroll or write. It’s the playlist that keeps the build focused and the ideas flowing.