AI/ML Engineer working across the modern AI stack, from LLMs, RAG, and agentic systems to generative AI, NLP, computer vision, and diffusion models.
Computer Science & Engineering student at BUET | Founder & AI/ML Engineer at Somokolon Labs
I am a Computer Science and Engineering student at BUET and an AI/ML engineer working across the modern AI stack, from LLMs, RAG, and agentic systems to generative AI, NLP, computer vision, and diffusion models. Alongside this, I research and fine-tune LLMs and turn them into production-grade systems built on fault-tolerant microservices and solid backend, full-stack, and DevOps engineering. To me, AI should be a core, observable part of production infrastructure, not just an add-on.
Founder & AI/ML Engineer
Jan 2026 – Present
Dhaka, Bangladesh
Multi-agent platform (plan -> research -> code -> critique) with RAG over documents; runs fully offline with local models.
Vector database built from scratch with HNSW/IVF indexes and AVX-512 distance kernels; Python bindings for search over millions of vectors.
On-device respiratory screening from cough audio; reimplemented preprocessing and ResNet18 in pure Dart, cutting the app from 600MB+ to ~70MB with no accuracy loss.
Offline-first outbreak-detection and medical-consultation system for low-connectivity rural regions of Bangladesh.
Future Builders 2025 (AI-Driven Cognitive Innovation Hackathon) — built HillTrack Pulse, an offline-first healthcare platform for rural Bangladesh.
INFINITE AI Builders (MarTech & Branding) — built AIgnis, an AI-driven autonomous marketing-campaign system.
BUET CSE Fest Hackathon — built OrionStream ML, an event-driven ML inference platform.
Bangladesh Physics Olympiad; National Participant in Physics & Astronomy Olympiads.
I write about machine learning, C++ backend optimizations, and full-stack development
A deep dive into how I built an offline-first multi-agent platform for code generation, research, and critique. Explores the system architecture, pgvector storage integration, and agent-critique loops.
Explores the math and engineering behind building a vector database from scratch. Shows how SIMD AVX-512 assembly speedups cut distance calculation latency.
How we cut on-device respiratory screening models from 600MB+ to ~70MB without losing accuracy, implementing preprocessing and neural net inference in Dart.
A research-focused blog detailing outbreak-detection algorithms used in HillTrack Pulse, enabling medical consultation in rural parts of Bangladesh.
Bachelor of Science in Computer Science & Engineering
July 2026
Higher Secondary Certificate, Science
June 2020
Notre Dame Science Club
BUET Robotics Club
Quantum Foundation Blood Donation Program
If you want to hire me, collaborate on AI/ML projects, or have other questions, drop me a message anytime!
shahriarseam17@gmail.com"If you don't take risks, you can't create a future."