Full-Stack Engineer & Problem Solver
Avizith Debnath
I am driven by solving real-world problems and understanding how things work under the hood. Expertise in building RAG systems and production-grade apps with FastAPI, Express, React, and Next.js.
Experience
AI Intern
Backend & AI Systems
Upscale BD Ltd.
2025 · 3 Months
- //Built production RESTful APIs with FastAPI — including a RAG system using sentence-transformers and Qdrant.
- //Optimized heavy processing tasks using Celery and BackgroundTasks, achieving sub-second latency.
- //Authored detailed Swagger/OpenAPI documentation to streamline team integration workflows.
Technical Proficiency
Languages
Web Tech
Databases
Tools
Featured Projects
Research Paper RAG Assistant
This modular RAG-based assistant streamlines academic research by automating the extraction of insights from dense PDF documents. Built with FastAPI, it utilizes Qdrant as a high-performance vector store for semantic retrieval and context-aware querying. To ensure system stability during heavy ingestion, I implemented semaphore-based concurrency controls for safe, synchronous file processing and optimized chunking strategies. The platform integrates Ollama for local LLM inference, providing a robust, privacy-focused solution for querying complex academic datasets with sub-second retrieval times and high factual accuracy.
Waste Management System — Code Samurai 2024
Developed for Dhaka North City Corporation, this smart logistics platform optimizes urban waste collection through advanced route planning and workforce management. As a winner at Code Samurai 2024, the system leverages the MERN stack and Google Maps API to visualize pickup points and generate efficient paths for collection vehicles. It features a citizen-facing reporting dashboard and a comprehensive administrative portal for contractor tracking. By digitizing manual reporting workflows and integrating real-time routing logic, the platform significantly reduces operational costs and improves city-wide disposal services.
StudyFlow — AI Educational Platform
StudyFlow is a comprehensive educational ecosystem designed to deliver personalized learning experiences through AI-driven content generation. Built with Next.js and TypeScript, the platform integrates Google’s Gemini AI to synthesize study materials, generate interactive quizzes, and provide real-time feedback. Recognizing the diversity of learning styles, I architected a progress-tracking engine backed by PostgreSQL to map knowledge gaps and suggest tailored study paths. This project was recognized as a finalist at Hack-The-AI 2025 for its innovative approach to bridging the gap between static curriculum and dynamic AI tutoring.
Hack-The-Future Platform
This full-scale hackathon management platform was engineered to simplify the complex lifecycle of developer events, from registration to final judging. It supports multi-stage judging pipelines, project submission workflows, and secure participant collaboration. I prioritized security and scalability by implementing a robust Role-Based Access Control (RBAC) system using Express and Node.js. The platform allows organizers to manage registrations, judges to score projects via custom criteria, and participants to showcase work in a centralized gallery. By automating logistical overhead, Hack-The-Future enables community leaders to focus on fostering innovation.
Pocket Tracker — Android App
Pocket Tracker is a performance-oriented personal finance companion built natively with Kotlin and Android’s Material Design components. Designed for professionals and students alike, the app allows users to log expenses, categorize spending, and gain actionable insights through intuitive data visualizations. Under the hood, it utilizes an offline-first architecture for fast data persistence, ensuring financial records are accessible without network dependency. The project focuses on clean UI/UX and efficient state management, helping users take control of their budgets by identifying spending patterns and optimizing savings through detailed analytical reports.
URL Shortener
Engineered as a high-concurrency backend service, this URL shortener provides a fast and reliable way to manage and track shortened links. Developed with TypeScript and Express, the system utilizes Redis for distributed caching to ensure lightning-fast redirects and minimal database load. I implemented a detailed analytics engine that tracks click-through rates, geographical data, and referral sources, providing users with actionable insights into link performance. The project demonstrates advanced backend patterns including rate limiting, input validation, and structured logging, making it a robust foundation for high-traffic production environments.
Achievements
Hack-The-AI Hackathon
Green University of Bangladesh
Code Samurai Hackathon
DU CSE collaborated with BJIT
Valid Certifications
B.Sc. in Computer Science & Engineering
Shahjalal University of Science and Technology
Expected Graduation
Blog
Insights Under the Hood
Writing about AI systems, backend architecture, and lessons learned while building in public. Currently documenting my journey with RAG pipelines and Full-stack scalability.