S. M. Sakeef Sani: medical AI that works with little data

Research Student · Kyushu University·Co-founder · Hinton Bit Corp

I build medical AI that works with little data and little compute.

Biomedical engineer from BUET. At Kyushu University I’m teaching speech AI to understand the dialects rural patients speak. I also teach AI and electronics as lectures you can play with, and tell stories in 3D.

0Citations0Awards0Courses taught

See the research Classes CV

01About

Four kinds of work. One question: who does it help?

Biomedical engineer from BUET, now a research student in Prof. Ashir Ahmed’s SocialTech Lab at Kyushu University. I build AI for healthcare where resources are scarce, teach at Northern University Bangladesh, co-founded Hinton Bit Corp and make 3D films.

I researchAI for healthcare Low-cost models for diagnosing mpox, dengue and heart disease, and now speech AI that understands rural Bangla dialects.
I teachLectures you can play with AI and digital electronics at Northern University Bangladesh, built as live, interactive lectures.
I buildVentures From agro-drones for Bangladeshi farms to enterprise systems at Hinton Bit Corp.
I create3D films & animation Award-winning shorts, VFX and a 3D animation course through my studio, Enigmatter.

02Research

Medical AI that works with little data. And understands how patients actually speak.

Now · SocialTech Lab, Kyushu University · Oct 2026 – Mar 2027

Speaking the patient’s dialect

The lab’s Portable Health Clinic brings checkups and telemedicine to communities in Bangladesh that hospitals don’t reach. But most Bangla speech recognisers learn from standard, read Bangla, and they stumble on the regional dialects rural patients speak.

I’m building a voice module that turns dialect speech into standard Bangla text and picks out symptoms for the health worker’s form. The data comes from Kontho (কণ্ঠ), a phone app I built that collects read speech for a parallel corpus of dialect, standard Bangla and English. Next: fine-tune Whisper and MMS, measure the dialect gap region by region, and shrink the model to run offline.

4,619: sentences in the corpus
4: regions: Rangpur, Narail, Barisal, Khulna
12: public speech and text datasets gathered

Kontho, the collection app → SocialTech Lab ↗

A Narail dialect sentence with its standard Bangla and English below

Recording the sentence, with a live waveform

Publications

0Citations0h-index0i10-index Google Scholar · Oct 2026 ↗ All research projects →

2022arXiv Co-author 331 citations
Monkeypox Skin Lesion Detection Using Deep Learning Models: A Feasibility Study

Introduced MSLD, one of the first public mpox skin-image datasets, during the 2022 outbreak, when PCR tests were scarce. Pre-trained models reached about 83% accuracy at telling mpox from diseases that look like it.
- Dataset
- Deep learning
- Mpox
arXiv: 2207.03342 →
2024Biomed. Signal Process. Control Co-author 141 citations
A Web-Based Mpox Skin Lesion Detection System Considering Racial Diversity

MSLD v2.0 adds patients across skin tones to reduce racial bias, benchmarks ResNet50 and other models with transfer learning from HAM10000, and ships a prototype web app for screening.
- Fairness
- Transfer learning
- Web app
Read on ScienceDirect →
2026CMPB Update Co-author 12 citations
BUET Multi-Disease Heart Sound Dataset: A Comprehensive Auscultation Dataset

BMD-HS: 864 multi-label heart-sound recordings covering complex valvular diseases, for training and testing computer-aided auscultation.
- 864 recordings
- Multi-label
- Valvular disease
arXiv: 2409.00724 →
2025AAAI Workshop First author
Efficient Antihallucinogenic AI for Tropical Medicine: A Probabilistic Framework for Differential Diagnosis

mLabLLM, my BUET thesis: a fine-tuned LLaMA 3.2 that reaches 82.8% top-3 accuracy on dengue, malaria and chikungunya, using Bayesian reasoning, LoRA and pruning to run cheaply. Presented at the AAAI 2025 workshop on LLMs and generative AI for health.
- LLaMA 3.2
- Bayesian
- LoRA
Read on OpenReview →

03Classes

Lectures you can play with. Literally.

I teach at Northern University Bangladesh. Every class is a live lecture with simulations, step-by-step solutions, quizzes and a student PDF. Try a piece of each course here.

Neuron CSE 4355

Artificial Intelligence & Expert Systems

Tap the hills to drop the climber.

8 of 20 classes live

Week 4: Informed search

Assignment 1: due Week 4 · Class 1

Assignment 2: due Week 5 · Class 2

Open CSE 4355 →

Volt CSE 2317

Digital Electronics & Pulse Technique

00 → 101 → 110 → 111 → 0

8 of 24 classes live

Week 4: Logic design I

Assignment 1: due Week 5 · Class 1

Open CSE 2317 →

04Work

Two research labs. Three ventures founded.

OCT 2026 — MAR 2027

Research Student

SocialTech Lab (Prof. Ashir Ahmed) · Graduate School of Information Science and Electrical Engineering, Kyushu University

Building a dialect-robust Bangla voice interface for the lab's Portable Health Clinic: Kontho, an app that crowdsources regional speech, and Whisper and MMS models fine-tuned to turn dialect speech into standard Bangla text. Supported by the Honda Foundation.

Kontho app →SocialTech Lab →
FALL 2026 — PRESENT

Adjunct Lecturer

Department of CSE, Northern University Bangladesh

Teaching CSE 4355 Artificial Intelligence & Expert Systems and CSE 2317 Digital Electronics & Pulse Technique. Every class is an interactive lecture: live simulations, step-by-step worked problems, assignments and class tests, all online for students.

All class materials →CSE 4355 · AI →CSE 2317 · Electronics →
2026 — PRESENT

Co-founder & Director of Operations

Hinton Bit Corp

Building custom and tailored solutions for government and private entities alongside some renowned global giants. Leading operations strategy and delivery at the cutting edge of enterprise technology.

hbitcorp.com →
2025 — 2026

Territory Officer

British American Tobacco

Managed a 420mn BDT annual distribution network. Developed an innovative model for rural 'Char' areas using satellite imagery, geographical data, and ROI analytics. Built an automated shipment app enhancing real-time route-to-market efficiency.
2021 — 2025

Research Assistant

mHealth Lab, BUET

Developed a real-time data collector using Raspberry Pi attachments. Contributed to a medical LLM for differential diagnosis. Created 3D anatomical landmarks from sparse 2D video data, boosting imaging accuracy in low-compute settings.

mHealth Lab →
2023 — 2024

Founder & CMLO

PakhiDrone BD

Led the creation of an agro-drone for precision farming and pesticide spraying, specifically adapted for Bangladesh's rural terrain and smallholder farming landscape. UIHP-funded and incubated into an early-stage startup.

05Awards

Johns Hopkins, Honda Foundation, Mars Society. Five awards, 2021 to 2024.

2024

Johns Hopkins Healthcare Design — Global Health Track

2nd Place · Team FetoSynth, BUET

Co-designed a real-time acoustic system for fetal heart sound localization and monitoring, improving prenatal care in low-resource settings.

View CBID Page →
2024

Johns Hopkins Healthcare Design — Digital Health Track

2nd Place · Team DengueDrops, BUET

Led Team DengueDrops with a smart IV fluid calculator for dengue patients to optimize fluid therapy and hospital management in endemic regions.

View CBID Page →
2023

Honda Young Engineer and Scientist (Y-E-S) Award

Honda Foundation · Bangladesh

Recognized for excellence in biomedical engineering and commitment to sustainable innovation. Flexible research grant supporting international exchange and ongoing research.

Honda Foundation List →YES Profile →BUET BME →
2023

University Innovation Hub Programme (UIHP) — Winner

BHTPA · PakhiDrone BD

Agro-drone project under PakhiDrone BD received UIHP funding and incubation support for precision farming technology in rural Bangladesh.

UIHP Program →Facebook Gallery →
2021

Innovation Award — International Planetary Aerial System Challenge (IPAS)

Mars Society South Asia · 1st Bangladesh · 8th Globally

Team Interplanetar (BUET) devised an innovative gas compression system for Mars aerial exploration — 1st in Bangladesh, 8th globally.

The Daily Star →Financial Express →Prothomalo →

06Studio

3D films, VFX and games. From my studio, Enigmatter.

Course · Enigmatter × MasterCourse

3D Animation MasterCourse

A full 3D animation course taught from my studio, Enigmatter, alongside some of the most iconic creatives in Bangladesh.

See the course ↗

Selected work

Enigmatter on Facebook ↗

07iOS & Blender

iOS apps and Blender tools. Built end to end.

I design and build iPhone apps for 3D capture and animation, and the Blender add-ons that bring their output into a real 3D pipeline.

Sanimate - Scan and Animate

iPhone · iOS 26 · Coming soon to the App Store

Scan your head in 3D. Animate it with your face.

An iPhone app that scans your head with the TrueDepth camera, wraps an animation-ready mesh onto it and drives it live with your own expressions, all on the device. Record takes and talking reels, stream to Blender, and export to OBJ, STL, PLY, USDZ or glTF with 52 blendshapes. The companion Blender add-on, Sanimate Link, imports the avatar with all 52 shape keys and drives it live from the phone over Wi-Fi.