Hi, I'm Raushan Kashyap

Autonomous multi-agent educational operations assistant for Ramakant Vidyapith. Tracks student attendance, CBSE scores, fee balances, and drafts instant bilingual (English/Hindi) WhatsApp progress alerts for parents.

• Domain Dispatcher routing queries across Registrar, Dean & Accounts agents
• Hybrid Engine: Local LLM (LM Studio / Ngrok) & Cloud Groq API integration
• One-click automated bilingual WhatsApp PTM notification generation

An autonomous customer service and ordering agent with conversational memory, multi-tool invocation for dynamic menu checking, live inventory tracking, and transactional write-backs.

• Stateful Multi-Turn Dialogue via MemorySaver
• Dynamic Tool Calling for FAQ & Stock Inquiries
• Real-time Transaction Execution & Inventory Writeback

Convolutional Neural Network (CNN) trained on handwritten digits, integrated with a live interactive HTML5 drawing canvas for real-time digit recognition.

• Feature extraction via Conv2D & MaxPool2D
• Interactive HTML5 Canvas Inference
• Real-time Probability Output Distribution

End-to-end regression pipeline analyzing structural feature vectors to estimate real estate valuations with high fidelity and zero variance leakage.

• Feature Engineering & Standard Scaling
• Cross-Validated XGBoost Optimization
• Interactive Valuation Input Dashboard

A decoder-only Autoregressive Transformer built ground-up in PyTorch, compiled to WebAssembly via ONNX, running 100% serverless inside client browsers.

• Multi-Head Attention & Causal Masking
• In-Browser Token Streaming with ONNX Runtime Web
• Instruction Fine-Tuned Dialogue Engine

An autoregressive Transformer model trained on MIDI pitch and duration tokens to compose and stream piano melodies in real-time directly inside the browser.

• On-Device WebAssembly Inference with ONNX Runtime Web
• Polyphonic Real-time Synthesis via Web Audio API
• Interactive Temperature & Token Sampling Controls