I'm a Computer Science student at Rangamati Science & Technology University currently in my 8th semester. I work as a Backend Tech Lead at FlowSurf.AI where I architect production RAG systems achieving 84.5% answer relevancy. I've solved 800+ competitive programming problems across Codeforces, LeetCode, and AtCoder, and ranked in the Top 30 out of 5,500+ participants at the FutureStack GenAI Hackathon. My research focuses on preventing temporal hallucinations in RAG using bi-temporal knowledge graphs.
Architected and deployed a production RAG system achieving 84.5% answer relevancy, 85.0% context recall, and 82.0% faithfulness (RAGAS) by building a tri-backend retrieval architecture with intelligent query routing and Cohere reranking, deployed as FastAPI microservices on GCP Cloud Run. Developed a production Model Context Protocol (MCP) server enabling AI agents to perform real-time read/write canvas operations.
Solved 800+ problems across Codeforces, LeetCode, AtCoder, CodeChef, and LightOJ. Ranked in the Top 30 out of 260+ projects among 5,500+ participants at FutureStack GenAI Hackathon.
From production RAG frameworks to real-time collaborative tools, here are some of the projects I've built and shipped.
Built an end-to-end Python framework integrating LightRAG, Graphiti, and Neo4j to construct bi-temporal knowledge graphs that track both event time and transaction time, with BFS-based multi-hop path exploration (hop decay, temporal edge scoring) and temporal intent detection via keyword/date-pattern parsing.
AI-powered platform that converts natural-language project ideas into complete PRDs, interactive mindmaps, and production-ready code scaffolding. Built enhanced MCP Gateway enabling multi-provider AI routing (Meta Llama, Cerebras, OpenRouter) with intelligent fallback.
Autonomous multi-agent research system built with FastAPI, Next.js & LangChain-style agent architecture that performs deep, iterative web research — decomposing queries into task graphs, validating sources with LLM-as-judge, extracting claims into a provenance evidence graph, and synthesizing cited reports via a reflexion self-correction loop.
Beyond building products, I contribute to open source, publish research on RAG architectures, and sharpen my skills through competitive programming — always pushing to learn and grow.
Academic Research
Proposed a bi-temporal knowledge graph architecture for Retrieval-Augmented Generation that addresses three failure modes — stale retrieval, cross-time contradiction, and chronological constraint violation — by fusing semantic similarity with temporal validity filtering across retrieval and multi-hop reasoning.
Online
Built the Deep Research Agent, an autonomous multi-agent research system with LangChain-style agent architecture that performs deep, iterative web research and synthesizes cited reports via a reflexion self-correction loop.
Online
Ranked in the Top 30 out of 260+ projects among 5,500+ participants, building AI-powered solutions with cutting-edge GenAI technologies.
Online Judges
Active competitive programmer solving problems across Codeforces, LeetCode, AtCoder, CodeChef, and LightOJ. Focused on data structures, algorithms, and problem-solving skills.
Open Source
Contributed 1000+ lines of code implementing responsive UI components including home banner, footer, navbar, and dark mode support improving 75% of the overall frontend. Merged 4+ PRs enhancing scalability and usability.
Want to chat? Feel free to connect with me on LinkedIn or drop me an email. I'm always open to interesting conversations and collaboration opportunities.
