Experience
Research, industry, academic, open-source, and leadership experience.
Research
FEB 2026 – PRESENT
Visiting Research Collaborator — Unsupervised RL for LLM Reasoning
- Designing intrinsic-reward signals to guide reasoning paths in large language models, mitigating mode collapse.
- Training reasoning agents with limited supervision for stronger self-correction in multi-step decision tasks.
- Building benchmarks to evaluate reasoning robustness and exploration coverage under distribution shift.
JUL 2025 – PRESENT
Independent Researcher — Debiasing & Explainability of VLMs
- Developing test-time debiasing methods for vision-language models to reduce demographic bias without retraining.
- Targeting medical and general VQA settings where labelled fairness data is scarce.
- Pairing the debiasing method with inference-time explainability traces.
2024 – PRESENT
Independent Researcher — Autonomous Driving with Efficient VLM Reasoning
- Trajectory and path planning for autonomous driving with explainable VLM-based reasoning.
- Using LoRA distillation on the Qwen open-source VLM with the Nexar dashcam dataset for real-time inference.
MAR 2024 – MAR 2025
Undergraduate Thesis Researcher
- Proposed MUSSE-Net: a multi-stage unsupervised sequential framework for ultrasound strain elastography.
- Designed a Contextual-Aware Feature Encoder, TriCrossAttention, and Cross-Attentive Sequential Decoder, implemented in PyTorch.
- Outperformed ReUSENet and re-implemented USENet on SNR, CNR, and NRMSE metrics, with improved lesion SNR.
- Validated on open-source simulation data, in-vivo benchmarks, and a clinical dataset.
Industry
APR 2025 – PRESENT
Machine Learning Engineer (Part-time)
- Medical-ExpertVLM — explainable VLM for personalised assistance to doctors and medical students.
- OCR & annotation pipeline using Qwen2.5-VL to extract medicine details and bank-cheque fields from handwritten images.
- AI-assisted system for drug discovery and development workflows.
Academia
JUN 2025 – PRESENT
Adjunct Lecturer, Department of CSE
- CSE428 — Image Processing: classical and deep-learning image processing for undergraduates.
- CSE350 — Digital Electronics: combinational and sequential logic design, lab and lecture instruction.
Leadership
UNDERGRAD
President — BUET Photographic Society (BUETPS)
- Led the campus-wide photographic society — events, exhibitions, workshops, and member onboarding.
- Coordinated photography for major university programs and represented BUETPS in inter-university events.
UNDERGRAD
President — BADHAN, BUET Zone
- Led the BUET zone of BADHAN — Bangladesh's largest voluntary blood-donation network.
- Organised blood-grouping camps, donor drives, and emergency blood-supply coordination across campus.