Hey, I'm Shiva Deepak.

An agentic RAG system that decides for itself whether to retrieve, ask for clarification, call a tool, refuse, or answer directly. Built with LangGraph for explicit state, three-type memory (conversation + episodic + semantic), and four retrieval modes with an honest ablation showing which actually helped.

• 7-node graph: decide → retrieve / clarify / tool / refuse / answer
• Three-type memory: conversation, episodic (LLM-summarized), semantic (user-profile)
• 4 retrieval modes benchmarked — vector, lightweight hybrid, true hybrid, cross-encoder
• 17 hand-written eval cases, scored on action accuracy + content match
• Honest result: cross-encoder didn't help. I said so in the README.

Monitors your PC for unauthorized access while long jobs run. Face recognition via OpenCV + DeepFace, Go WebSocket backend, Next.js dashboard, and a rules engine for process/CPU/network triggers. Alerts via Discord, Telegram, and email.

• v0.6.0 — actively maintained with versioned releases
• Multi-channel alerts: Discord · Telegram · Email · Windows toast
• Rules engine: process, CPU, network, disk, schedule triggers

Hadamard gates generate truly random cryptographic keys; XOR + pixel permutation encrypts the image. Includes full statistical analysis — entropy, histogram uniformity, correlation, and PSNR metrics — with a Streamlit interface.

• Hadamard-gate quantum key generation
• Hybrid XOR + permutation cipher
• Statistical security analysis built in
• Perfect lossless reconstruction on decrypt

A proposed standard for tracing what an AI coding agent actually did — file-level changes, decisions, and reasoning — so the next agent (or human) can audit it.