I am a software engineer focused on building high-throughput backend architectures, applied GenAI / RAG pipelines, and low-latency network systems. My engineering approach emphasizes robust API design, caching layers, database indexing, and writing clean, scalable code.

• 🏎️ Backend & Systems: Deep experience with FastAPI, Node.js/Express, C++17, PostgreSQL, Redis, and MongoDB, designing RESTful APIs and multithreaded network engines.
• 🤖 Applied AI & RAG: Built production-grade Retrieval-Augmented Generation workflows with LangChain, SentenceTransformers, and FAISS, leveraging Redis caching to drop inference query latency from 2.5s to under 15ms.
• 💡 Algorithmic Problem Solving: 750+ DSA problems solved, Knight on LeetCode (Rating 1943, Top 3.5%), 3-Star on CodeChef (Max Rating 1639), and Pupil on Codeforces (1246), with contest finishes in the Global Top 285 on LeetCode and Round 2 Qualifier in Google The Big Code 2026.

Platform	Achievement / Max Rating	Global Standing / Highlights
LeetCode	Knight (Max Rating 1943)	Top 3.5% globally · Global Rank 285 (Weekly Contest 502) · Global Rank 1,488 (Biweekly 184)
CodeChef	3-Star (Max Rating 1639)	Handle: itss_levi
Codeforces	Pupil (Rating 1246)	Handle: aryangupta_
Google The Big Code	Round 2 Qualifier (2026)	Top 1,500 nationwide qualifier
Problem Solving	750+ Problems Solved	Strong foundations in Trees, Graphs, DP, Sorting/Searching, Flow Networks
JEE Main 2023	Top 3% Nationwide	Out of 1.2+ Million Candidates
Leadership	Captain, BIT Mesra Football Team	(2025 – Present) Leading squad strategy, fitness, and team coordination

AI / RAG · FastAPI · PostgreSQL · Redis · FAISS · LangChain · React · Docker
🔗 GitHub Repository · 🎥 Video Demo Walkthrough

• End-to-End RAG Pipeline: Ingests multi-format documents (PDF, DOCX, TXT) with LangChain recursive chunking, generates embeddings with HuggingFace SentenceTransformers, and performs top-$k$ semantic retrieval using an in-memory FAISS vector store.
• FastAPI REST Backend: Built with Pydantic validation, SQLAlchemy ORM, and PostgreSQL for document metadata, query logs, and analytics.
• Multi-Turn Conversational Q&A: Maintains session-aware conversation context, streams responses to the React UI in real-time, and asynchronously triggers 3 smart follow-up suggestions per answer.
• High-Performance Caching: Integrated Redis query caching (1-hour TTL) and rate limiting, drastically dropping repeated query latency from 2.5s down to under 15ms (~99.4% latency reduction). Fully containerized with Docker & Docker Compose.

[User Document (PDF/DOCX/TXT)] ──► [LangChain Chunker] ──► [SentenceTransformers] ──► [FAISS Vector Index]
                                                                                              │ Top-k Chunks
[User Query] ──► [Redis Cache Check] ──(Miss)──► [FastAPI Backend] ──► [LLM + Context] ───────┘
                     │ (Hit: <15ms)                                      │
                     └──────────────────── Streamed Response ────────────┴──► [React UI + Follow-Ups]

C++17 · PCAP Parsing · Multithreading · TLS SNI · 5-Tuple Flow Tracking · CMake
🔗 GitHub Repository

• High-Throughput C++17 Core: Co-engineered an offline Deep Packet Inspection engine from scratch with zero heavy external dependencies, featuring a custom PCAP reader and Ethernet, IPv4, TCP, UDP, and ICMP header parsers.
• Multithreaded Architecture: Utilized thread-safe queues and producer-consumer pipelines capable of processing 10,000+ packets/sec on a single CPU core.
• Traffic Classification: Extracts Server Name Indication (SNI) from TLS 1.2/1.3 Client Hello packets, decodes HTTP Host headers, and resolves DNS queries to accurately identify web and application protocols.
• 5-Tuple Flow Tracking & Filtering: Built a bidirectional stateful flow tracker (src_ip, dst_ip, src_port, dst_port, protocol) with a rule engine that filters traffic by IP, wildcard domains, applications, and ports, exporting filtered PCAP captures and HTML audit dashboards.

[.pcap Capture] ──► [PCAP Reader] ──► [Ethernet / IPv4 / TCP / UDP / ICMP Parsers]
                                                        │
                                            [Thread-Safe Queue]
                                                        │
                      ┌─────────────────────────────────┴─────────────────────────────────┐
                      ▼                                                                   ▼
       [TLS SNI / HTTP Host / DNS Classifier]                              [5-Tuple Stateful Flow Tracker]
                      │                                                                   │
                      └───────────────────────────────┬───────────────────────────────────┘
                                                      ▼
                                           [DPI Rule Engine]
                                   (Block by IP / Domain / App / Port)
                                                      │
                            ┌─────────────────────────┴─────────────────────────┐
                            ▼                                                   ▼
                [Filtered .pcap File]                              [JSON / CSV / HTML Dashboard]

React.js · Node.js · Express.js · MongoDB · JWT · AWS S3 · AWS EC2 · PM2 · TanStack Query
🔗 GitHub Repository · 🌐 Live Demo

• Full-Stack MERN Architecture: Engineered a responsive social platform featuring user authentication, post creation, comments, likes, following relationships, and user profile management.
• Stateless JWT Security: Implemented JWT authentication stored in HTTP-only cookies to eliminate XSS-based token theft vulnerabilities.
• Algorithmic Feed Personalization: Implemented a custom HackerNews-inspired time-decay scoring formula: $$\text{Score} = \frac{(\text{Likes} \times 1) + (\text{Comments} \times 5)}{(\text{AgeInHours} + 2)^{1.5}}$$
• $O(1)$ Read Latency via Cron Worker: Deployed an asynchronous node-cron background job that periodically precomputes trending scores in MongoDB, ensuring lightning-fast feed retrieval regardless of post volume.
• Cloud Infrastructure: Integrated AWS S3 for scalable media storage, TanStack React Query for intelligent client-side caching, and deployed the production backend on an AWS EC2 (Ubuntu) instance with PM2 for process clustering and zero downtime.

Node.js · Express.js · MongoDB · EJS · Cloudinary · Render
🔗 GitHub Repository · 🌐 Live Demo

• Full-stack accommodation booking and rental listing application built with Node.js, Express, MongoDB, and EJS templating.
• Includes user registration, authentication, full CRUD listing management, review & rating systems, and Cloudinary media upload integration.

May 2026 – June 2026 · Remote

• Engineered core full-stack features for Soul Consciousness Lab using React.js, Node.js, Express.js, and MongoDB.
• Built production-ready RESTful APIs with strict request validation, structured error handling, and robust CRUD capabilities.
• Profiled frontend components to eliminate redundant API invocations, minimize unnecessary React re-renders, and implemented code-splitting with lazy loading to significantly boost page load speeds.

I am actively looking for Software Engineering (SDE), Backend, Full-Stack, and AI/ML internship roles. If you have an opening or would like to discuss engineering and collaboration, please reach out!