Selected build work
Software systems, research prototypes, and tools I built while trying to understand how the code actually behaves.
Projects
OngoingAn ongoing Rust project that turns repository tasks into compact, evidence-labeled context plans and packs for Codex, Claude Code, Cursor, OpenCode, and other MCP-compatible coding agents. v1 now includes real-client MCP smoke proof for...
Research
An in-depth exploration of applying "Design by Contract" principles to Large Language Models, offering a comprehensive taxonomy, detection methods, and enforcement strategies to make AI systems more reliable and predictable.
Systems Programming
A comprehensive implementation of a custom database engine in Go featuring ACID compliance, B+ tree storage, MVCC concurrency control, adaptive indexing, and ML-enhanced query optimization - built entirely from scratch without external database libraries....
Research
An advanced blockchain framework designed specifically for healthcare data management that incorporates sharding, Layer-2 solutions, and a Directed Acyclic Graph (DAG) ledger structure.
Projects
A blockchain ticketing prototype for fraud-resistant issuance, controlled resale, and verifiable entry. Built as a BCOLBD 2021 finalist project.
Freelance learning project
A freelance learning project where I am building a small Docker-like runtime in Go: process isolation, namespaces, filesystem setup, container networking, and command-line behavior.
Freelance learning project
A freelance learning project where I am rebuilding the basic parts of PyTorch: scalar values, tensors, autodiff, neural-network layers, convolutional operators, and simple training loops.
Projects
A deep dive into ray tracing implementation, exploring the physics of light transport, rendering equations, and practical optimization techniques for creating photorealistic 3D graphics.
Projects
A custom convolutional neural network implementation from scratch using only NumPy that achieves 95.87% accuracy on Bangla handwritten digit recognition, without relying on deep learning frameworks.
Projects
Extended the "Show, Attend, and Tell" image captioning model with BERT to improve caption quality and reduce training time, achieving 36% improvement in CIDEr score and 43% improvement...
Projects
A comprehensive microservices-based event management platform engineered for scalability, resilience, and rapid feature delivery, supporting end-to-end workflows for event creation, promotion, participant engagement, and monetization.
Projects
Modified TCP Vegas implementation with dynamic aggressiveness to address fairness issues when competing with TCP Reno, achieving near-perfect fairness while preserving Vegas' efficiency advantages.
Projects
A sophisticated multi-pass compiler for a subset of C programming language, featuring lexical analysis, syntax analysis, semantic analysis, intermediate code generation, and assembly optimization.
Ongoing Rust project for packing useful repository context for coding agents. The current version can inventory a repo, retrieve lexical and symbol matches, include dependency and git-history signals, build Markdown or JSON context packs, and run MCP smoke tests with Codex CLI and Claude Code.
Freelance learning project where I am building a small Docker-like runtime in Go. It covers process isolation, namespaces, filesystem setup, container networking, and command-line behavior.
Freelance learning project where I am rebuilding the basic parts of PyTorch: scalar values, tensors, autodiff, neural-network layers, convolutional operators, and simple training loops.
Finalist in Blockchain Olympiad Bangladesh (BCOLBD) 2021 with team "Recursively Enumerable". Designed NFT-based ticketing system using ERC-1155 standard. Implemented smart contracts for anti-scalping, dynamic QR codes, and decentralized identity management with zero-knowledge proofs.
Implemented Show, Attend and Tell architecture with ResNet-101 encoder and LSTM decoder. Achieved BLEU-4: 0.335, CIDEr: 0.92 on MS-COCO dataset. Enhanced with beam search and multi-head attention achieving 11-point BLEU-4 improvement. Conducted comprehensive ablation studies and attention visualizations.
Designed microservices-based event management system. Led back-end architecture implementing newsfeed, payment, authentication, and event management services.
Built complete compiler for subset of C language using Flex (lexical analysis) and Bison (parsing). Implemented symbol table management with scope handling and comprehensive error reporting. Added semantic analysis for type checking and function validation.
Implemented and analyzed TCP congestion control variants (Reno vs Vegas) using NS3 network simulator. Designed TCP Vegas+ modification addressing fairness issues through dual-mode operation. Conducted comprehensive performance analysis measuring throughput, fairness index, and packet drop ratios.
