Passionate dev: full-stack, AI/ML, Systems programming

codeforces

leetcode

github

Jishnu Prasad  • Peerlist

Cross-platform application to retrieve and process browser history locally, enabling natural language queries over previously visited webpages through Retrieval-Augmented Generation Developed cross-platform application to retrieve and process browser history locally Implemented natural language queries over previously visited webpages through RAG Built Python FastAPI backend with React + TypeScript frontend Utilized ChromaDB for vector storage and Ollama running LLaMA 3.2 for intelligent content recall Created custom embedding pipeline for browser history data Implemented real-time query processing with streaming responses

Browser History Local RAG App

Completely free and open-source, cross-platform markdown-based notes application with real-time preview and advanced features Built cross-platform markdown-based notes application with real-time preview Implemented hierarchical folder organization, auto-save, and full-text search Designed custom markdown renderer with interactive checkboxes and dark/light theming Added share-intent integration and offline-first local persistence Used optimized React Context state management for performance Implemented custom syntax highlighting for code blocks Implemented custom whiteboard with export and import fucntionality

Ingenium (Markdown Notes App)

Redis-compatible in-memory data store with RESP support, custom hash-table storage, and advanced features Built a Redis-compatible in-memory data store with RESP support Implemented custom hash-table storage, goroutine-based concurrency, TTL expiration, and snapshot persistence Added natural-language sorting, case-insensitive commands, direct key access, and clearer error messages Created accompanying GUI and CLI clients for better user experience Implemented data persistence with periodic snapshotting Added comprehensive test coverage for all commands

Memora (In-Memory Data Store)

Designed a system that avoids traditional messaging channels by leveraging WebRTC encoded insertable streams to hide packetized payloads inside outbound video frames and reconstruct them on the receiving end. Key contributions and features: Built secure room-based peer-to-peer communication using WebRTC Implemented a WebSocket signaling server for SDP, ICE, and key exchange Designed covert data transfer by embedding encrypted payloads within encoded video frames Implemented end-to-end encryption using ECDH key exchange and AES-GCM Developed real-time packetization, embedding, extraction, and reconstruction pipelines Added live call metrics including FPS, latency, and payload overhead Architected a modular system with React (Vite) frontend and Node.js backend, integrating cryptographic workflows with real-time media pipelines. This project explores the intersection of real-time communication, cryptography, and steganography, demonstrating how media streams can be repurposed for secure and hidden data transmission.

Project CipherStream

An experiment to determine ability of LLMs to control a rover and guide it from one point to another using GPS and continuously updated context of routes in disaster situations Developed hybrid AI + rule-based navigation system using YOLOv8 (vision) + ultrasonic sensing for fallback incase of hihg latency Integrated LLM-based control via OpenRouter and Ollama running on edge with constrained decision outputs and intelligent fallback on failures (rate limits, timeouts) Engineered real-time WebSocket pipeline (ESP32 <-> Raspberry Pi 5) for sensor streaming and low-latency control Implemented median filtering + consecutive detection logic, achieving ~96% obstacle avoidance reliability Designed fail-safe architecture with timeout handling, graceful degradation, and motor safety constraints Optimized for edge deployment (Raspberry Pi 4) with efficient CPU (25–45%) and memory usage (~480MB)

Project Nomad

Designed and implemented a system that combines real-time packet capture with process-level attribution, enabling users to identify which applications generate network activity. Built a dual-interface experience: • A structured flow table for detailed inspection • A force-directed graph visualization where network hosts are represented as dynamic nodes sized by traffic volume Integrated Linux-based process mapping via /proc, along with a lightweight HTTP API for exposing flow data and enabling external integrations. Developed an MCP (Model Context Protocol) integration that allows natural language querying of network activity, enabling users to retrieve insights such as top traffic sources, protocol usage, and process-specific flows. This project emphasized rethinking traditional network monitoring by making traffic patterns more interactive, visual, and easier to interpret.

Project PacketLens | IET NITK Project Expo 20

Engineered low-latency AIS ingestion pipeline (WebSockets + Node.js) handling 1000+ vessels with ~50–200 ms latency Designed hybrid streaming + RAG architecture integrating Redis (live state) and ChromaDB (historical context) Reduced data noise by ~95% via dual-layer geospatial filtering (bounding box + Haversine radius) Implemented vessel position prediction + exponential smoothing to stabilize sparse, noisy GPS signals Built FastAPI + React system with real-time WebSocket broadcasting for multi-client visualization Enabled natural language queries over live + historical vessel behavior (e.g., anomaly detection, proximity analysis) Optimized for high-throughput, low-latency processing with scalable in-memory state management (Redis TTL)

RescueMind - Real-Time Disaster Intelligence 

A locally-first Kanban-style productivity desktop application with a maritime theme and neo-brutalist design, built for managing complex projects, notes, and tasks with optional cloud synchronization.

Shipyard

Full-stack system monitoring solution built with React frontend and Django backend, providing real-time monitoring, alerting, and management of distributed monitoring agents Developed full-stack system monitoring dashboard with real-time agent and host metrics Implemented secure agent registration workflow with centralized management Built alerting system for multi-level security and resource alerts Integrated JWT authentication and role-based access control for users Visualized CPU, memory, disk, network usage, and live process data Implemented encrypted communication between agents and server Created modular React components with lazy loading and protected routes Backend built with Django REST framework and PostgreSQL database