# Ashraful Kabir Alif

> AI researcher and ML engineer in Dhaka, Bangladesh (BRAC University; AI Developer at Shahriar Labs). Builds machine learning that holds up when the signal is faint and the stakes are high: exoplanet detection in JWST imaging, kidney tumour/stone segmentation, safe offline RL for sepsis treatment, variant effect prediction, ADMET prediction, and in-browser computer vision tools.

## Research
- Exoplanet direct imaging (JWST/MIRI): contrastive, few-shot Siamese ResNet18 — 92.16% accuracy. BSc thesis; paper submitted.
- MoE-RenalSAM: SAM2 + LoRA mixture-of-experts for renal tumour (Dice 88.19%) and stone (Dice 84.06%) segmentation: https://github.com/AshrafulKabir7/MoE-RenalSAM
- Action-masked conservative Q-learning for sepsis (ICU-Sepsis benchmark): https://github.com/AshrafulKabir7/sepsis-offline-rl
- VariantFusion (MSc thesis, in progress): calibrated, isoform-aware variant effect prediction.
- Multi-task GATv2 for 23 ADMET endpoints: https://github.com/AshrafulKabir7/ADMET_Research

## Publications
- Pattern Recognition in Disaster Response: Leveraging Machine Learning for Twitter Analysis. ICITA 2024, LNNS (Springer), 2025. https://doi.org/10.1007/978-981-96-1758-6_4
- Submitted (2025): exoplanet detection with contrastive learning on JWST data; monotonic neuro-symbolic semantic loss; explainable chromatin accessibility prediction.
- In preparation: fair and explainable multi-disease chest X-ray diagnosis.

## Builds
- Defecto — in-browser anomaly detection for visual inspection: https://defecto-five.vercel.app
- Tallycam — in-browser detection, tracking and counting: https://tallycam.vercel.app
- Citely — grounded document Q&A with citations: https://citely-pi.vercel.app
- CaseCraft — legal drafting with hybrid RAG: https://github.com/AshrafulKabir7/PSL-Flow

## Contact
- Email: md.ashraful.kabir.alif@g.bracu.ac.bd
- GitHub: https://github.com/AshrafulKabir7
- LinkedIn: https://www.linkedin.com/in/ashrafulkabiralif
- ORCID: https://orcid.org/0009-0001-1011-1236