Applied intelligence, shipped.
AI/LLM Engineer building production NLP and generative systems inside a regulated bank. Creator of BRACBankGPT, the first internal generative AI assistant in Bangladesh's banking sector. I work across the whole path — fine-tuning, retrieval, inference, and the automation that makes it pay for itself.
- 0
- Years shipping AI
- 0+
- Production bots
- 1st
- Gen-AI assistant in BD banking
Where I’ve been
building.
01 / Experience
Applied AI inside a regulated bank — where the constraint isn't the model, it's everything around it.
2023 → NOW
Associate Manager, R&D (AI and ML)
BRAC Bank PLCProcess Automation and R&D · DhakaDesign and ship production AI systems inside a regulated bank — generative assistants, document intelligence, and email automation — from fine-tuning through inference and deployment.
Selected
projects.
02 / Case Studies
Most of this runs inside a bank, so the screenshots are limited and the links are private. Happy to walk through any of it.
01 / 06
At BRAC Bank PLC · Generative AI Assistant
BRACBankGPT
Archived case study · walkthrough by request
A secure, internally hosted ChatGPT-like assistant supporting employees with fast, general-purpose responses while keeping data inside the bank and within regulatory bounds. Enabled safe AI usage without reliance on external tools.
- Role
- Design and build, end to end
- Impact
- First of its kind in BD banking
- Craft
- Self-hosted inference, RAG grounding, full data privacy
02 / 06
At BRAC Bank PLC · Document Intelligence
Multimodal LLM Fine-Tuning
Archived case study · walkthrough by request
Fine-tuned a multimodal LLM (Qwen3-VL) on 750 hand-annotated domain-specific documents to raise document understanding and extraction accuracy, guided by base-model failure analysis. Deployed with vLLM for low-latency inference across OCR and document intelligence pipelines.
- Role
- Fine-tuning and deployment
- Impact
- Higher extraction accuracy over base model, low-latency serving
- Craft
- LoRA/PEFT via Unsloth + TRL SFTTrainer, vLLM serving
03 / 06
At BRAC Bank PLC · Document Intelligence
Document AI Extraction Pipeline
Archived case study · walkthrough by request
Asynchronous Qwen3-VL extraction API for Credit Card Onboarding and Account Opening Forms, replacing a legacy Microsoft Document AI system. Extracts and validates form data with a mandatory human review step before anything reaches core banking systems.
- Role
- Design and build, end to end
- Impact
- ~200 documents/day at 90–95%+ accuracy, 40–55 seconds per document, replaced Microsoft Document AI
- Craft
- Structured JSON output, human-in-the-loop review, full audit trail
04 / 06
At BRAC Bank PLC · Process Automation
Guarantor Link Update RPA
Archived case study · walkthrough by request
Orchestrated ~50 RPA bots reading validated data from an internal database and entering it into an external regulator site, with secure Active Directory authentication and daily MIS reporting.
- Role
- Solo build, bot orchestration and infrastructure
- Impact
- ~105 hours/day saved, ~16 months of manual work eliminated across 15 staff (~9.36M BDT)
- Craft
- AD-authenticated bot fleet, daily automated reporting
05 / 06
At BRAC Bank PLC · Email Automation
LLM-Based Email Classification
Archived case study · walkthrough by request
A dual-layer email automation system combining keyword-based rules with a semantic similarity model, so one layer compensates when the other misclassifies. Runs 24/7 including holidays, with automated document delivery and customized replies.
- Role
- Solo build
- Impact
- 186+ hours saved per month
- Craft
- Embedding-based classification, dual-layer redundancy
06 / 06
Personal · Portfolio Experience
This Site
A statically exported Next.js build treating the resume as a product surface, with its own design system and an AI assistant answering questions about the work above.
- Role
- Solo design and build
- Craft
- Next.js static export, custom design system, AI chat assistant
The stack,
honestly.
What I actually reach for — not everything I’ve ever opened.
Education
Year- B.Sc. Computer Science and Engineering — BRAC University
- 2023
Certification
Year- Advanced Certificate for Management Professionals (ACMP) 4.0 — IBA, University of Dhaka
- 2025
Fellowship
Year- Perplexity AI Business Fellowship — Perplexity
- 2025
Generative AI & LLMs
- Retrieval-Augmented Generation (RAG)
- Prompt Engineering
- LoRA / PEFT Fine-Tuning
- Embeddings
- Transformers
- Self-Hosted LLM Serving (Ollama, vLLM)
- LangChain
- LangGraph
- LlamaIndex
AI-Native Engineering Tools
- Claude Code
- Codex
- MCP (Model Context Protocol)
- n8n
Machine Learning
- PyTorch
- TensorFlow
- Hugging Face
- Sentence-Transformers
- Semantic Similarity Search
- Model Evaluation and Benchmarking
- Dataset Annotation and Curation
Document AI / Computer Vision
- OCR / ICR
- Qwen3-VL OCR
- GLM-OCR
- Surya
- Chandra
- YOLO
- Bilingual Bangla + English OCR
Vector Search / Retrieval
- FAISS
- Pinecone
- Chroma
Programming Languages
- Python
- JavaScript
- TypeScript
- SQL
- Bash/Shell
- Java
- Go
- Rust
Backend & APIs
- FastAPI
- Asynchronous REST APIs
- SOAP Integration
- Redis
- JWT Authentication
- Role-Based Access Control (RBAC)
Databases
- MS SQL Server
- SQLite
- PostgreSQL
- MySQL
- Oracle
Infrastructure & Cloud
- Docker
- On-Premises GPU Deployment (NVIDIA L40S)
- Microsoft Azure
- Azure AI Foundry
- Nginx
- Git
04 / Get in touch
