Connect an AI model

Use your own key for free-form answers. Without a key, the built-in guide answers from the portfolio.

Voice uses Gemini Live with a soft female voice, in Bangla or English. Without a Gemini key, your browser’s own voice is used.

Your key stays in this browser and is sent only to the provider you pick. Unless you tick “Remember”, it’s forgotten when you close the tab.

Mubin Ul Islam Chowdhury

AI Engineer, Researcher & Trainer

4+ years building production AI at BRAC IT Services, doing research that holds up to measurement, and teaching engineers to do the same.

  • EngineerMulti-agent platforms, RAG, biometrics
  • ResearcherIEEE-published, with an MIT Astrobee scientist
  • TrainerExperienced trainer of engineering teams
Email me See my work

Open to research positions and AI engineering roles · Ask my AI assistant

Mubin Ul Islam Chowdhury

4+ yearsIndustry AI R&D

1st Runner-UpCode Sprint 2026

100% Rank-1FVC2002 benchmark

Worked with

  • BRAC IT Services
  • BRAC University
  • STEMX365 with NASA/MIT
  • JAXA Kibo-RPC
  • IBA, University of Dhaka

Engineer, researcher, trainer.

4+ years of applied AI R&D: shipping systems, publishing research, and training engineering teams.

I’m an AI Engineer, Researcher and Trainer with 4+ years of industry R&D at BRAC IT Services, working across machine learning, computer vision, and NLP. I build enterprise AI, such as the multi-agent BRAC IT AI Kit, RAG systems, and biometric identification at scale, alongside research prototypes measured with information theory and statistical estimation.

I co-authored peer-reviewed work on space robotics with an MIT Astrobee scientist through STEMX365. I hold a B.Sc. in Computer Science and Engineering from BRAC University and ACMP-4.0 from IBA, University of Dhaka, and I’m looking for research positions and engineering roles in AI.

  • IEEE XploreCo-author of ISS repair-mission research with an MIT Astrobee scientist, 2023
  • Mars rover leadLed 17 engineers on BRACU Mongol-Tori, 2022
  • Hackathon podiums1st Runner-Up, Code Sprint 2026; 2nd Runners-Up, AI Engineering Hackathon 2025
  • Best MentorBRAC IT, 2026, for three training sessions

Selected work.

R&D and competition work at BRAC IT Services, 2025–2026. Most of these systems are proprietary, so what follows is what they do and how well.

BRAC IT AI Kit

A production, self-hosted multi-agent chatbot platform.

An LLM “selector” router dispatches each query to specialist agents across sequential, tree, graph, and hybrid workflows, with a Swagger/OpenAPI importer and 50+ tools spanning business APIs, web search, and RAG.

50%+ lower repeat-query LLM cost, from multi-layer Redis caching.

  • 16 pre-built agents, plus full Procurement and University Admission assistants
  • Provider-agnostic through LiteLLM: cloud models, or local vLLM and Ollama
  • Keycloak and RBAC authentication
  • Prometheus and OpenTelemetry observability
  • Python
  • FastAPI
  • CrewAI
  • LiteLLM
  • React 19
  • PostgreSQL
  • Redis
  • Kubernetes
~15k LOC · Open-source edition: Delaxis

Advance Fingerprint Indexing

Large-scale biometric identification and deduplication.

A retrieve-then-rerank cascade: vector and LSH retrieval narrows the gallery, then an exact geometric matcher decides. A ten-finger fusion gate returns an auditable DUPLICATE, CLEAN, or REVIEW verdict, with scale analysis to 10 million records.

100% Rank-1 on FVC2002, and 100% recall@10 at up to 5001× gallery reduction.

  • Java 25
  • SourceAFIS
  • Qdrant
  • PostgreSQL
  • MinHash-LSH

TenderSense

An AI SaaS that tells a business which tenders to bid on.

Scheduled pipelines aggregate live tenders from multiple procurement platforms. A bid-advisory engine matches each company profile against the specification and returns Bid or No-Bid, with email alerts and Admin and Super Admin panels.

48 hours from brief to shipped product. 1st Runner-Up of 33 build-round teams.

Team Neuro Warriors, with Arnab Dey and Lita Mouri Sarker · BRAC IT Code Sprint 2026

DataAssistant

Conversational data visualization that cannot make numbers up.

A find–compute–draw architecture separates semantic retrieval, validated read-only SQL, and deterministic chart binding. Text and live-voice modes share one tool surface, with role-based access and full query audit logging.

Every figure traces to a real database row.

  • Python
  • FastAPI
  • Angular
  • Gemini
  • sqlglot
  • PostgreSQL
  • MCP
  • Keycloak

Cricket Entropy Analysis

Information-theoretic analytics for ball-by-ball cricket data.

Batter predictability through Shannon entropy and exploitability through mutual information, with bias-corrected estimators and bootstrap intervals. A vision pipeline cuts broadcast video into per-delivery clips.

Small-sample noise is never presented as signal.

  • Python
  • DuckDB
  • NumPy
  • YOLO26
  • OpenCV
  • Ollama VLM

Live siteSource

Weft

Context, woven: one grounded answer from documents, spreadsheets, and databases.

Hybrid retrieval fuses dense search and BM25 by reciprocal rank. A context graph links every chunk to its section, neighbours, and entities, and personalised PageRank walks it, so one question can hop from a PDF into a database table. Numeric questions become read-only SQL whose result is a citable passage.

It refuses when the evidence is thin. Confidence combines retrieval strength, question coverage, source agreement, and citation rate. It is not the model’s self-report.

  1. Sources
  2. Chunker
  3. Contextualizer
  4. Embedder + context graph
  5. Retriever
  6. Reranker
  7. Answer
  8. Guardrail
  • Python
  • React
  • Qdrant
  • pgvector
  • FAISS
  • BM25
  • PageRank
  • MCP server
55 unit + 13 end-to-end tests · Proprietary · BRAC IT

More work

  • Bangla Document RAGOllama-hosted Gemma retrieval-augmented generation for Bangla documents.Source
  • Put It OnVirtual try-on with OpenCV and Gemini 2.0 Flash image generation.Source
  • SQL GeneratorNatural language to SQL, built three ways: Java with LangChain4j, Python, and N8N.
  • Flowable Modeler GeneratorLLM-generated Flowable process models, with SpringBoot and LangChain4j.
  • BRAC University USISThe university management system, full-stack in Angular and SpringBoot.
  • PostgreSQL fingerprint extensionBiometric matching inside PostgreSQL and PHP, written in C/C++ with SourceAFIS as a reference.
  • Online exam monitoringComputer-vision proctoring built on OpenCV.
  • Enterprise chatbotsHRMS and procurement assistants with automated agent routing in N8N.

Open source.

Projects of my own, outside BRAC IT. Delaxis and Iron Base are public, as is Cricket Entropy Analysis above.

Where I’ve built and shipped.

Roles, teams, and the systems that came out of them.

  1. Software Engineer (R&D) & AI Researcher

    BRAC IT Services Ltd., Dhaka

    Oct 2022 – Present Now

    Agent platforms

    The BRAC IT AI Kit, a self-hosted multi-agent platform with a visual workflow builder, and an Ollama-based RAG system for complex documents.

    Enterprise automation

    A Flowable Modeler Generator, scalable SQL generators, and HRMS and procurement chatbots with automated agent routing.

    Full-stack and backend

    The BRAC University Management System (USIS), a Quarkus two-factor-authentication service, and a reactive notification system on Spring WebFlux.

    Systems and teaching

    A C/C++ fingerprint extension for PostgreSQL and PHP, and the internal “AI-Powered Development” workshop for engineering teams.

  2. Space Technology & AI Integration Researcher

    NASA/MIT & STEMX365 Collaborative Research

    2022 – Present Now

    • Collaborated directly with an Astrobee scientist (MIT) on AI for autonomous navigation and free-flying robotics aboard the International Space Station.
    • Contribute to cross-disciplinary R&D across AI, mechanical robotics, and space science for operational simulations and international educational outreach.

  3. Team Lead, BRACU Mongol-Tori

    Semi-autonomous Mars rover · BRAC University

    Jan 2022 – Oct 2022

    • Directed the AI development lifecycle for a team of 17 engineers.
    • Real-time object detection (YOLOv5, SSD MobileNet) across 9 rock types, and OpenCV pathfinding with Arrow and ArUco marker detection.

  4. Workshop Facilitator & Campus Representative

    BRAC University

    2019 – 2022

    • Organised and delivered workshops on artificial intelligence, neural networks, and Python for data science, mentoring students in deep-learning methods.

Research and teaching.

Published work, and the workshops I run for engineering teams.

Predicting Suicidal Intent from Social Media Text Posts Using Machine Learning

Machine-learning models that flag suicidal intent in social media text posts.

  • Machine learning
  • NLP
  • Mental health

M. U. I. Chowdhury · DSpace, BRAC University · 2022

AI-Powered Development: Enhancing Coding with Intelligent Integration

I led BRAC IT’s internal workshop, upskilling engineering teams. Three comprehensive training sessions earned the 2026 Best Mentor Award.

  • AI tooling
  • Developer experience

Instructor · BRAC IT Services

Stack and foundations.

Languages, frameworks, mathematics, and the credentials behind the work.

AI & Machine Learning

  • Deep Learning
  • Computer Vision
  • NLP
  • RAG
  • LLMs
  • Multi-agent systems
  • Predictive Modeling
  • MCP

Programming

  • Python
  • Java
  • C/C++
  • Go
  • JavaScript
  • TypeScript
  • SQL

Frameworks & Libraries

  • TensorFlow
  • PyTorch
  • scikit-learn
  • OpenCV
  • LangChain
  • Transformers
  • YOLO
  • SpringBoot
  • Angular
  • React

DevOps & Infrastructure

  • Docker
  • Kubernetes
  • GCP
  • Git
  • CI/CD
  • Kafka
  • RabbitMQ
  • Redis
  • PostgreSQL
  • Qdrant
  • DuckDB

Mathematics

  • Calculus
  • Probability
  • Linear Algebra
  • Statistics
  • Information Theory

Certifications

  • DataCamp: Data Science & Machine Learning Specialization; Statistics; Supervised & Unsupervised Learning
  • Coursera: Programming for Everybody (Python)
  • HackerRank: Problem Solving

Education

B.Sc. in Computer Science and Engineering
BRAC University · 2018–2022 · CGPA 3.54 / 4.00
ACMP-4.0
IBA, University of Dhaka · 2025
HSC (Science)
Collectorate School and College, Rangpur · GPA 5.00 / 5.00

Let’s talk.

Open to research positions and engineering roles in AI.

Email me LinkedIn