Open to opportunities

Shahariar Rabby

I turn hard ML problems into ships — medical imaging deployed across 21 hospitals, LLMs serving 30,000 users a day.

Research Scientist · UAB / Senior ML Engineer · Apurba Technologies
45+
Publications
1,000+
Citations
17
h-index
30K
DAU served
0.92
COPDxNet AUC

Research

PhD dissertation: Precise Segmentation of Thin, Elongated, and Underrepresented Structures (UAB, 2026) · CS Outstanding PhD Student Award · Advisor: Dr. Chengcui Zhang

Vision & Document AI
Computer Vision
Thin structure segmentation, medical imaging, infrastructure monitoring, OCR pipelines. Edge deployment at 29.6 FPS on a Raspberry Pi 5.
Applied AI & Analytics
Decision Support
Long-tail learning, model calibration, evaluation in high-stakes domains.
Deployable AI (MLOps)
Systems
Scalable serving, GPU inference optimization, latency-aware deployment.
Responsible AI
High-Stakes Domains
Privacy-aware clinical & enterprise AI, human-in-the-loop validation.

Selected Publications

2025

STRIPE: Directional Feature Extraction for Pulmonary Fissure Segmentation In preparation

Manuscript in preparation

First Author · Dice 0.873 · HD95 0.72mm · AFM + DDA · LCNet-100

2025

THREAD: Efficient Topology-Preserving Network for Thin Structure Segmentation In preparation

Manuscript in preparation

First Author · 48.6% fewer FLOPs · 92 FPS GPU / 29.6 FPS Pi 5

2024

Enhancement of Bengali OCR by Specialized Models and Advanced Techniques

IEEE/CVF WACV Workshops, 2024

First Author

Engineering

Shipped systems that run in production today — government platforms, banking AI, and a RAG pipeline acquired by Promise.

Kagoj.ai
Bangladesh Government AI Platform
Fine-tuned Qwen2.5 VLM via LoRA with FP8 quantization and Flash Attention 2, deployed on vLLM with L40S GPUs. 30,000 daily active users · 1–2K peak RPS.
vLLMVLMLoRAFP8
FinDoc
Enterprise Document Processing
End-to-end AI platform for Bangla+English banking documents. OCR, layout analysis, extraction at 90%+ accuracy. 5 years of continuous development.
OCRDocument AIBanking
Dorsal.fyi Acquired
RAG Pipeline → Promise, Feb 2026
Architected a HIPAA-compliant RAG pipeline for a health-tech startup, acquired by Promise.
Borno
Bangladesh National OCR Platform · ocr.bangla.gov.bd
Multi-stage CNN OCR on Triton + ONNX: 18.6× speedup, 10K+ daily requests, 99.5% uptime. Live →
TritonONNXCNN

Project details

Selected systems, with simplified workflows and reported evaluation results.

Document AI · Production platform

Kagoj.ai

Bangladesh Government AI Platform

Fine-tuned Qwen2.5 VLM using LoRA, FP8 quantization, and Flash Attention 2, served with vLLM on a single NVIDIA L40S GPU.

View system details

Illustrative document-AI workflow

Conceptual flow only; platform request-handling components are not specified here.

  1. 01Document inputDocument image or page
  2. 02Qwen2.5-VLDocument understanding
  3. 03Structured outputModel response

My role & implementation

Fine-tuned the vision-language model with LoRA and optimized inference using FP8 quantization and Flash Attention 2. Deployed model serving through vLLM on one NVIDIA L40S GPU.

Reported scale

30,000 daily active users and 1–2K peak requests per second are platform-wide figures. They describe platform traffic, not GPU inference throughput.

Medical imaging · Multicenter evaluation

COPDxNet

COPD detection from chest CT

A lightweight 3D CNN evaluated using COPDGene development data and external cohorts.

View system details

Conceptual evaluation workflow

Simplified study flow; this is not a deployment diagram or an exact model architecture.

  1. 01Chest CTStudy scans
  2. 023D CNNCOPDxNet model
  3. 03Model outputDetection score
  4. 04Cohort evaluationInternal + external

My role

Led architecture design and multicenter evaluation for COPDxNet.

Reported evaluation

13,043 COPDGene CT scans from 21 centers. AUC: 0.92 standard-dose and 0.88 low-dose COPDGene scans; external AUC: 0.92 on SPIROMICS and 0.82 on NLST.

Manuscript under review at The Lancet Digital Health.

Grants

NIH/NHLBI R01 HL151421
Research Scientist
"Structural Determinants of Disease Progression in COPD" — PIs: Surya Bhatt, MD & Arie Nakhmani, PhD. Research Scientist since Jun 2026; GRI 2022–2026.
NSF SaTC #2154589 · $397,302
2022–2027
"Collaborative Research: Bubble Aid" — ballot tabulation research. PI: Chengcui Zhang.

Experience

Research Scientist Current
UAB — CLAIR & KDDM Lab
Jun 2026 – Present

NIH R01: Fissure Integrity Score — quantitative pulmonary fissure metric for COPD phenotyping. Lead developer of COPDxNet and STRIPE.

Senior ML Engineer Current
Apurba Technologies
Nov 2025 – Present

Led a 13-engineer team. Qwen2.5 VLM on Kagoj.ai (vLLM, L40S). Event-driven microservices on Kafka. Dorsal.fyi RAG pipeline.

Graduate Research Assistant & Blazer Fellow
UAB — CLAIR & KDDM Lab
Aug 2021 – Apr 2026

PhD: Precise Segmentation of Thin, Elongated, and Underrepresented Structures. COPDxNet, STRIPE, THREAD, dCrack61K. NSF SaTC: ballot tabulation.

ML Engineer → Team Lead
Apurba Technologies
Oct 2019 – Aug 2021

Borno (ocr.bangla.gov.bd). Triton + ONNX: 18.6× speedup. Grew the team from 3 to 10+, 10K+ daily requests, 99.5% uptime.

Lecturer (Part-Time)
Daffodil International University
Jan 2019 – Dec 2020

Taught AI and AI Lab. Shifted curriculum to project-based learning.

Awards

CS Outstanding PhD Student AwardUAB · 2026
Barrett R. & Oceana A. Bryant Endowed AwardUAB · 2024
Best PaperIEMIS · 2023
Best Paper — "Ekush"RTIP2R · 2018
Blazer Fellowship · CVPR Travel GrantUAB · 2022
3-Minute Research Pitch — 1st Runner UpSBCL · 2022
Best Student Research AwardDIU · 2018 & 2019
IBM Centre of Excellence MVP2020–2021

Datasets & Artifacts

Mentoring & Service

Mentored 13 undergrads
To First-Author Publications
Research Supervisor, DIU NLP & ML Lab since 2019. Multiple mentees now pursuing PhDs. WVLL @ WACV 2024 organizer; WVLL 2026 lead organizer at UNT.
33 peer reviews
Top Venues & Journals
IEEE TDSC (9) · IJMDEM (6) · IEEE TMM (3) · ACM Multimedia (6) · WACV (4) · ICMLA (3) · AAAI (1) · CVPR (1). Webmaster, UAB Graduate Student Government 2024–25.

Let's build something
worth deploying.

Research collaboration, AI systems work, or just a conversation.