The Student Brain Engine.
SNUGPT is the institutional intelligence layer built explicitly for the Shiv Nadar University (SNU, Delhi-NCR) ecosystem, empowering students with immediate, unified knowledge retrieval.
Institutional Vision
University campuses are complex, highly regulated micro-ecosystems. Between fragmented ERP portal pages, institutional handbooks, hostel rules, exam procedures, and academic pre-requisites, students are often forced to manually hunt through dozens of PDFs and web pages to find simple answers.
SNUGPT was engineered by students, for students, to serve as a unified, conversational knowledge layer. By merging highly structured institutional context with cutting-edge Large Language Models, SNUGPT acts as an instantaneous, neural query center that answers questions with extreme precision in seconds.
Technical Architecture & RAG Specs
Inference Layer
Powered by Meta Llama 3.1 8B Instruct running over high-speed NVIDIA AI Foundation endpoints, streaming answers token-by-token for commercial-grade speed.
Vector Database
Semantic lookup uses highly advanced nv-embedqa-e5-v5 embeddings indexed inside a secure Chroma vector store to fetch exact policies from scanned documentation.
Concurrency Scaling
FastAPI backend is optimized with thread-safe double-checked lock patterns (`threading.Lock`) and recycled PostgreSQL connection pools supporting up to 100 concurrent channels.
Persistent Memory
Uses highly resilient serverless Neon PostgreSQL connected via asyncpg client pools for secure, private user session history management.
Integrated Ingestion Pipeline
To provide accurate, context-grounded answers without hallucination, SNUGPT integrates multiple knowledge pipelines:
- ERP Academic Catalogs: Detailed course credits, departmental codes, prerequisites, and registration procedures.
- SNU Student Handbooks: Essential legal regulations, grading curves, CGPA thresholds, and disciplinary policies.
- Student Welfare Guidelines: On-campus hostel rules, curfews, medical support directories, and event protocol guides.
Every source is cleaned, chunked into overlap-aware passages, converted into vector representations, and safely cached to ensure retrieval queries respond with exact-match citations.
Rishabh Joshi Creator
SNUGPT was conceived, designed, and fully engineered by **Rishabh Joshi** as an open-source contribution to the Shiv Nadar University community. It aims to showcase the power of localized, retrieval-augmented intelligence interfaces in solving everyday institutional friction.