Selected Projects
Graph-Based Timing Prediction at Early-Stage RTL Using Large Language Models
- Developed an RTL dataset generator to produce thousands of synthetic circuits, creating a diverse graph dataset.
- Converted gate-level Verilog into a DAG capable of handling millions of nodes and utilized LLemma-7B to generate node and graph-level embeddings from technology files, which were in a mathematical format.
- Implemented a GAT to enhance graph representation, leveraging graph-level features after global mean pooling and before passing them to the linear layer.
Python
PyTorch
Graph Neural Networks
Large Language Models
Verilog
Generalizable and Relation Sensitive Netlist Representation for Analog Circuit Design
- Proposed a novel heterogeneous graph representation for analog circuits.
- An embedding lookup table for each component type in the circuit with weights learned through backpropagation.
- Introduced a novel RL reward method to optimize the unmet metric when other metrics are met.
- Solved the sample inefficiency issue, providing better generalization and enhanced knowledge transferability across different specifications and circuit topologies, and outstanding performance in zero-shot settings without further training.
Python
PyTorch
Reinforcement Learning
Graph Neural Networks
Analog Circuit Design
RFID Based Smart Print Zone for NSU
- Designed and developed an RFID-compatible printing system for centralized print zones to handle client requests.
- Enabled users to print documents from home, offering a cost-effective alternative to conventional printing solutions.
- Actively used by more than 30,000 students and faculty members at NSU.
Acquired by NSU for 2 million BDT
PHP 7.4
Laravel 8
jQuery
MySQL
RFID
Webapp for Impulse Hospital and a Full-Fledged HMS
- Worked on a hospital management system, which encompasses accounting and billing services, prescription services, online physician services, human resource management, enhanced reporting system management, and an online pharmacy.
Currently used by over 1,000 healthcare professionals and staff
PHP 8
Laravel 8
RESTful APIs
React
MySQL
RyoGas: A SaaS-Based Web Application
- Developed a web application providing IoT solutions for automating the remote management of fuel stations.
- A SaaS architecture to manage accounting, gas price adjustments, bank statements, gas stock, and monthly reports.
- Integrated a POS app to record gas sales and print receipts.
Currently in active use across more than 550 gas stations
PHP 7.4
Laravel 8
jQuery
MySQL
IoT
SaaS
POS