Subash Saajan J

• Designed and built the platform's AI grading system, which marks student assignments, quizzes and discussions and lets teachers review every grade before it is released.
• Rebuilt the AI assistant that handles tasks across the platform as a multi-step agent on LangGraph, with long-term memory and tools limited to what each user's role allows.
• Built the hybrid RAG behind the assistant, combining vector and keyword search so answers come from each course's own material, and it declines when the sources don't support an answer.
• Built an evaluation harness that runs simulated teacher and student conversations and scores the replies with an LLM judge, to catch drops in answer quality.
• Ship features across the whole product, from the database and backend APIs to the React/TypeScript app on AWS, including hardening API access control.

• Customers scan a QR code, upload a document, pay by UPI and collect the print, either at a self-service kiosk or at a partner print shop.
• Wrote the backend in Go (Fiber) with PostgreSQL, Redis and Cloudflare R2 for file storage, and the customer site in Next.js.
• Wrote the device software in Go, a Raspberry Pi agent for kiosks and a Windows app for print shops, both pulling jobs from the server.
• Each job's state is saved before it reaches the printer, so a crash mid-print is caught on restart and a stuck job can move to a backup printer.
• Updates are signed. Devices check the signature and file hash before installing, wait until nothing is printing, and hold no passwords or API keys.
• CI builds the Raspberry Pi and Windows versions for every release. The server runs on Docker Compose behind Nginx with TLS.

• Led the rebuild from scratch in TypeScript, covering the React Native Android app, NestJS backend, admin console and website.
• Built the web admin console where staff manage bookings, home visits and payments, and the Razorpay payment integration.
• Set up hosting and deployments on Google Cloud, with checks that test every change before it goes live.

• An assistant built on LangGraph and Gemini that drafts emails and calendar events, and the user confirms before anything is sent.
• Long-term memory on PostgreSQL with pgvector, and Gmail and Calendar access through Google OAuth.
• Was deployed on AWS (ECS Fargate, RDS) with Docker, Terraform and GitHub Actions.

Springer CCIS, presented at ASCIS 2024 (Marwadi University, Rajkot). A blockchain-enhanced geographic routing framework to strengthen location authentication and mitigate spoofing in VANETs. Awarded Best Paper. DOI: 10.1007/978-3-031-86302-8_25