Intelligence R&D Lab
We work alongside students, researchers and organisations on academic research, theses, publications and software development, with a dedicated mentor guiding every engagement from scope to delivery.
Measured on held-out test data
99.43%99.43%
Brain tumor classification from 7,023 MRI scans, EfficientNet-B1Brain tumor classification, 7,023 MRI scans
120,000120,000
Real and AI-generated images used to train a deepfake detectorImages used to train a deepfake detector
0.96 AUC0.96
Best class in a five-way Bangla cyberbullying classifierAUC, best class in a Bangla cyberbullying classifier
8,8928,892
Skin images in an arsenicosis dataset built from field photos in ChapainawabganjSkin images in a field-built arsenicosis dataset
01 — Selected work
A few recent projects across medical imaging, computer vision, NLP and energy. Client names and author details are kept private.
Recent projects across medical imaging, computer vision, NLP and energy. Client names and author details are kept private.
Swipe to see all 8 projects
Six CNN architectures trained on 7,023 MRI scans to tell glioma, meningioma, pituitary tumors and healthy brains apart. Grad-CAM and SHAP show which regions drove each prediction, and the best model runs in a Flask and Streamlit web app.
99.43%test accuracy
Seven models, from classic CNNs to a Vision Transformer and SimCLR, benchmarked on 120,000 real photos and Stable Diffusion fakes. Every verdict comes with a Grad-CAM map of the pixels behind it.
98.22%with DenseNet
Medical imaging · Dataset
1,482 photos from four villages in Chapainawabganj, augmented to 8,892 and used to compare supervised and self-supervised models with Grad-CAM.
NLP · Bangla
Sorts comments into five categories, compares classical models with BanglaBERT, and ships as a Flask and MongoDB web app for moderators.
Health analytics · XAI
Machine learning and ensemble models on 70,692 CDC survey records with 22 features, explained with LIME so each prediction can be checked.
Computer vision · Industry
Routes each item to recyclable, compostable or landfill. Five CNN architectures compared and the sorting line simulated end to end.
Computer vision · Smart home
Lightweight CNNs benchmarked for on-device occupancy detection; MobileNetV2 reached 99.35% test accuracy.
Energy · Data analysis
Data analysis, model development and simulation for a renewable-powered industrial data center, with tuned models compared side by side.
Medical imaging · XAI
Six CNNs on 7,023 MRI scans, with Grad-CAM and SHAP explaining each prediction, deployed as a web app.
Computer vision
Seven models benchmarked on 120,000 real and Stable Diffusion images, each verdict mapped with Grad-CAM.
Medical imaging · Dataset
1,482 field photos from Chapainawabganj, augmented to 8,892 to compare supervised and self-supervised models.
NLP · Bangla
Five-way classifier comparing classical models with BanglaBERT, shipped as a Flask and MongoDB web app.
Health analytics · XAI
ML and ensemble models on 70,692 CDC survey records, explained with LIME.
Computer vision · Industry
Recyclable, compostable or landfill: five CNNs compared and the sorting line simulated end to end.
Computer vision · Smart home
Lightweight CNNs benchmarked for on-device occupancy detection.
Energy · Data analysis
Data analysis, modelling and simulation for a renewable-powered industrial data center.
02 — Services
Start from a blank page or bring a half-built project. Each engagement gets a mentor, a clear scope and a timeline agreed before work begins.
Not sure what you need? Message usScope, build and document your project with a mentor working alongside you or your team.
Research design, literature review, methodology and analysis, worked through chapter by chapter.
Turn finished work into a conference-style paper with clean figures, result tables and references.
Structured technical reports with system diagrams, Gantt charts, budgets and clearly presented results.
Already building? Get reviews, debugging help and direction from people who have shipped similar work.
Cleaning, exploration, statistics and modelling, with charts you can stand up and explain.
03 — What you get
Plus the working code, trained models and a deployed demo where the project calls for one.
Plus the working code, trained models and a deployed demo where the project calls for one.
04 — How it works
Tell us your topic and deadline in a short form, as a team or on your own.
We call within 24 hours to agree scope, timeline and fee, then open your project.
A mentor is assigned and the work moves through development and review, with you in the loop.
You receive the full set of files. Payments are logged against your project until the balance clears.
05 — Fields we work in
Our AI projects pair a strong model with explainability, so you can show reviewers why it works, not only that it does. Our software projects are built as real, deployable web and mobile apps, designed and documented the way industry teams do it.
06 — Questions
Anything else? Write to [email protected].
Yes. Solo students register exactly the way teams do; choose "Just me" on the form.
It depends on the scope and your deadline. We agree the fee on the first call, before any work starts, so there are no surprises.
No. You pay in instalments through bKash, Nagad, bank transfer or cash, and every payment is recorded against your project.
Often, yes. Alongside AI and ML we take on data analysis, engineering and energy projects, theses and report writing. Message us your topic and we will tell you honestly whether we are the right fit.
Yes. The work shown on this page has client names and author details removed, and we never share your files without permission.
Anything else? Write to [email protected].
Fill in the registration form and we will call you back within 24 hours.