Hi, my name is
Munawar Mahtab Ansary.
I build intelligent AI systems.
I'm a Machine Learning Engineer specializing in Natural Language Processing, AI systems, and full-stack development. Currently, I'm focused on building innovative ML solutions at Innospace Infotech Ltd., working on projects involving RAG, computer vision, and conversational AI.
About Me
Hello! My name is Munawar and I'm passionate about building intelligent AI systems that solve real-world problems. My journey in machine learning began during my undergraduate studies at BRAC University, where I developed a deep interest in Natural Language Processing and low-resource language technologies.
Fast-forward to today, and I've had the privilege of working at Innospace Infotech Ltd. as a Machine Learning Engineer, DataSoft as an intern, and BRAC University as a Student Tutor. My main focus these days is building RAG-based chatbots, computer vision systems, and NLP pipelines that enhance user experiences and automate complex workflows.
I recently completed my undergraduate thesis on BanglaMUX, a project focused on enhancing regional dialect detection, transcription, and translation for low-resource Bangla language using transformer-based approaches.
Here are a few technologies I've been working with recently:
- Python
- PyTorch
- TensorFlow
- Django
- Laravel
- RAG
- NLP
- AWS
- MySQL
- Git

Where I’ve Worked
Machine Learning Engineer @ Innospace Infotech Ltd.
September 2025 - Present
- Shipped a production RAG chatbot (live on Brritto) that answers user questions with context-aware, retrieval-grounded responses
- Built a GPT-OSS-20B + Qwen3-VL vision-language QA system that detects and rectifies errors in uploaded questions before publication
- Built an OCR-to-structured-content pipeline that converts scanned exam papers (image or text) into exam-ready question sets for the Brritto app
- Engineered Innovert, a legacy-font-to-Unicode conversion system for a national publisher, and deployed it to production on AWS EC2 with systemd
- Built AI agents for automated communication over Facebook and WhatsApp using n8n
Some Things I’ve Built
Featured Project
QuickTrips - Car Rental Platform
Developed a car rental web application that suggests optimized trips based on user preferences. Integrated chatbase for intelligent recommendations, implemented secure authentication, and connected multiple APIs to provide a seamless booking experience.
- Django
- HTML
- CSS
- Chatbase
- REST APIs
Featured Project
Sexism Detection with BiLSTM
Built a Bidirectional LSTM model to classify and categorize online sexist content using the SemEval 2023 EDOS dataset. Preprocessed imbalanced real-world text data and implemented GloVe embeddings for binary classification, achieving robust performance on detecting harmful content.
- Python
- Deep Learning
- BiLSTM
- NLP
- GloVe
- TensorFlow
Featured Project
BanglaMUX: Regional Dialect Detection & Translation
Undergraduate thesis project focused on developing a transcription and translation model for Bangla dialects using transformer-based approaches. Built a pipeline that takes regional dialects of Bangla language and converts them into textual standard Bangla. Explored various tools and models including Whisper, BERT, Bangla T5, and mT5 for low-resource language processing.
- Python
- Transformer Models
- Whisper
- BERT
- Bangla T5
- NLP
Featured Project
DeshScan — Private On-Device Document Scanner
A free, private document scanner where your documents never leave your device. Photograph any document — NID card, certificate, form — and get a clean, flattened, shareable scan. A compact U-Net, trained on self-generated synthetic shadow pairs, removes shadows and color casts, and the whole pipeline (detect → dewarp → shadow removal → OCR → PDF) runs locally, even on CPU. Supports Bangla + English OCR.
- Python
- PyTorch
- U-Net
- OpenCV
- OCR
Featured Project
MonCare — Trilingual Mental-Health Support Chatbot
A private, trilingual (Bangla · Banglish · English) mental-health support chatbot that runs entirely on one local GPU. It detects the user's language, screens every message through a two-stage crisis gate, and answers with retrieval-augmented generation over a 19K-item counseling knowledge base (bge-m3 embeddings, FAISS, cross-encoder rerank, local LLM). Crisis responses are deterministic and route to verified human helplines, with an output guard that strips any helpline number the model invents. Built as a Research Engineer Intern at UIU's AIMS Lab. Not a medical device.
- Python
- RAG
- FAISS
- GPT-OSS-20B
- FastAPI
Other Noteworthy Projects
view the archiveAuto Fund Mailer
A web app that helps students find professors matching their research background and reach out for funded PhD/Masters positions. It reads the student's CV, ranks professors by research fit, and writes personalized funding-request emails that are sent from the student's own Gmail.
Amar Kotha — Bangla AAC App
A free, open-source Bangla AAC (Augmentative and Alternative Communication) app that gives non-verbal children a voice. Children tap picture tiles and the app speaks a full, grammatical Bangla sentence. It pairs a deterministic grammar engine (the safety-critical path) with a fine-tuned BanglaT5 neural composer trained on self-generated, order-shuffled data — reaching near-perfect exact match on tile combinations it never saw in training.
Microfin Department Chatbot
Developed a full department-based chatbot for Microfin360 using Retrieval-Augmented Generation (RAG) to provide accurate responses to department-specific queries during internship at DataSoft.
Airborne Autos - Obstacle-Jumping Car Game
Developed a 2D obstacle-avoidance game where players control a car that jumps over dynamic obstacles like trees and stones. Implemented collision detection, physics-based jump mechanics, and progressive difficulty using Python and OpenGL.
Flight Management System
Designed and implemented a web-based flight booking platform supporting both one-way and round-trip searches with advanced filtering options. Integrated secure authentication, automated e-ticket issuance, and invoice generation. Developed an admin dashboard to manage flights, routes, fares, and passenger records.
What’s Next?
Get In Touch
Although I’m not currently looking for any new opportunities, my inbox is always open. Whether you have a question or just want to say hi, I’ll try my best to get back to you!
Say Hello

