Experience, education & skills

Experience

1. Machine Learning Engineer, Accelx Inc.

   Jan 2025 — Present

   Real-time video understanding, vision-language models and grounded LLM systems · Dhaka, Bangladesh

   Violence recognition in video

   • Fine-tuned a Kinetics-pretrained R(2+1)D-18 and deployed it for real-time multi-camera monitoring.
   • Compared it on RWF-2000 with optical-flow, pose-fusion and pose-Transformer models, and it beat all three.
   • Also fine-tuned a YOLOv10-L weapon detector for live streams.

   Checking what the pose models learned

   • Traced their inter-person "contact" cues to keypoints missing for both people, and found the models did not track people across frames, so could not tell who did what.

   Zero-shot fall and violence recognition with CLIP

   • Built recognizers on a frozen OpenCLIP model, traced their false positives on real footage to camera viewpoint, lighting and posture, and turned these cases into hard-negative prompts, a fix that needs no retraining.

   Grounded LLM and VLM outputs for human reviewers

   • Built a pipeline where a local LLM writes alerts only from facts computed in code and fine-tuned RoBERTa risk scores, and added an automatic check on every citation.
   • Designed an OCR → CLIP → Qwen3-VL cascade that calls the VLM only for hard cases, checks findings against a knowledge base and flags weak evidence for review.
   • Built zero-shot URL categorization with BART-MNLI and Redis.

Education

1. B.Sc. in Computer Science and Engineering

   Apr 2016 — Jan 2021

   Ahsanullah University of Science and Technology · Dhaka, Bangladesh

   • Thesis: Design of an Arrhythmia Classification Algorithm Using 2-D Convolutional Neural Networks
   • Relevant coursework: Artificial Intelligence, Pattern Recognition, Digital Image Processing, Computer Graphics, Soft Computing, Data Structures, Algorithms

Technical skills