FSE_PantherBot

AI-powered academic advising platform for Chapman University's Fowler School of Engineering

Overview

FSE_PantherBot is a sophisticated RAG (Retrieval-Augmented Generation) system that provides 24/7 academic advising support via the Fowler School of Engineering Slack Workspace. It combines multiple AI technologies to deliver personalized, accurate guidance to undergraduate students using university catalogs and policies.

Architecture

Core Components

• Document Ingestion: Apache Tika extracts content from PDFs and academic catalogs
• Embedding Model: Qwen3-Embedding generates semantic embeddings for documents and queries
• Vector Database: Qdrant stores document embeddings with hybrid search capabilities
• Reranker: BGE-Reranker-V2-M3 improves retrieval relevance
• LLM: Qwen3.5: 30B generates final responses via Ollama
• Memory System: PostgreSQL stores conversation history with automatic compression

Retrieval Pipeline

1. Query Router: Combines semantic similarity and LLM-based routing to select appropriate document collections
2. Hybrid Search: Fuses dense (semantic) and sparse (BM25) retrieval for optimal results
3. Reranking: Refines retrieved chunks using cross-encoder model
4. Response Generation: LLM synthesizes answers using retrieved context and conversation memory

Memory Management

• Conversation Storage: PostgreSQL tracks user interactions and chat history
• Memory Compression: Intermediate LLM summarizes conversations to maintain context while reducing tokens
• Student Profiles: Persistent storage of major, catalog year, and academic preferences

Configuration

All system parameters are configured in configs/config.yaml
See configs/README.md for detailed options including:

• Model selection and parameters
• Retrieval weights and thresholds
• Memory compression settings
• Collection-specific configurations

Usage

Local Deployment

• needs to be updated

DGX0 Compute Cluster

# Sync code to cluster
./scripts/sync_to_cluster.sh

# Deploy on cluster assuming ssh keys are set up properly
ssh dgx0.chapman.edu
./scripts/run_cluster.sh -b -c -f

Make an ssh tunnel (assuming hostname is already configured; else update that)

ssh -L 10001:localhost:10001 -L 10002:localhost:10002 dgx_cluster

Run Streamlit

streamlit run src/streamlit_app.py

I need to update this README because the documentation is bad

Run docker compose up -d to start up containers (QDrant, PostgreSQL, 2 Ollama Containers)

Run Ingestion python fse_ingestion/ingest.py