Automated License Plate Recognition / Registration System

The ALPR system, an epitome of cutting-edge technology, utilizes Optical Character Recognition (OCR) on images for precise extraction of vehicle registration data. Driven by GRPC for seamless client-server communication, it integrates YOLOv3 and YOLOv4 models for meticulous object detection and a MongoDB NoSQL database for efficient data storage. With a character-level accuracy of 93%, the ALPR system stands as a robust solution for real-time vehicle tracking and registration.

An image search and image-text matching system for Bangla

An image search and image-text matching system for Bangla using CLIP (Contrastive Language–Image Pre-training). The model consists of an EfficientNet / ResNet image encoder and a BERT text encoder and was trained on multiple datasets from Bangla image-text domain. The image search engine for the Bangla language will take a text query and from an image database, it will return the most relevant k image results.

Ramanujan - LLM Search Engine for Developer Cloud

An LLM Search Engine - that can find developers with specific attributes (skills, location, hourly rate, education tier, company tier, industries, seniority level, etc.) through natural text search from a pool of 3 million developers. My contribution included (but not limited to): prompt engineering for extracting job information, writing, running, and scheduling all the ETL pipelines for performing developer attribute extraction periodically, fine-tuning text-bison model from Vertex AI for quality improvement, fine-tuning and deploying Llama 13-B for cost optimisation, evaluation of the LLM results with human ratings, etc.

Automated Speaker Recognition Engine (ASRE)

The TASRE project boasts advanced features for speaker recognition. It supports 1 to 1 and 1 to N matching, incorporating proprietary techniques for silence detection, noise removal, and speech feature extraction. The system offers robustness, scalability, and easy integration, with live and offline enrollment options. Its technical specifications include a tensor-based file format for efficient embedding vector storage. Experimental results showcase impressive performance on diverse datasets, emphasizing its text-independent nature.

Explaining Trees with FastTreeShap and What if tool

The What If tool is an interactive explainability plugin from the People + AI Research (PAIR) Initiative of Google. The What-If Tool (WIT) offers a simple user interface for understanding regression and black-box classification ML models better. The plugin allows you to execute inference on a big collection of samples and view the results right away in a number of different ways. Here's a PoC with What If Tool and FastTreeSHAP to explain trained tree models before productionaisation.