PedalBot: AI Research Assistant for Guitarists

Project Description

Guitarists often face the challenge of navigating long, complex effect pedal manuals when trying to find specific information quickly. PedalBot solves this by acting as a research assistant for guitar pedals.

It’s a Retrieval-Augmented Generation (RAG) application that allows musicians to upload effect pedal manuals or query preloaded ones and receive instant, LLM-powered answers with cited sources. This ensures accuracy and saves time, making it easier to explore features, troubleshoot, or compare pedals.

Core Functionality and Architecture

RAG Pipeline

PedalBot implements the full RAG flow required in Module 1:

1. Document Ingestion

   • Supports PDF uploads (manuals, guides).
   • OCR support for scanned PDFs using Google Vision API.
   • Extracted text is stored with metadata (source, page).

2. Embeddings & Retrieval

   • Uses VoyageAI embeddings.
   • Stores and retrieves documents via Pinecone vector database.

3. LLM Generation

   • Answers are generated with Groq LLM.
   • Responses include source citations and page numbers for traceability.

4. Guardrails

   • Always follows the prompt configurations from the Yaml file and links answers back to original documents to minimize hallucination.

5. User Experience (UX)

   • FastAPI backend handles ingestion, retrieval, and LLM pipeline.
   • Streamlit frontend provides an intuitive interface to upload manuals and query them.

Technical Stack & Setup

• Languages & Frameworks: Python, FastAPI, Streamlit
• Vector Store: Pinecone
• Embeddings: VoyageAI
• LLM: Groq
• OCR: Google Vision API
• Deployment: Render (backend), Streamlit Cloud (frontend)

Deployment Links

• Backend (FastAPI): https://research-assistant-oe9n.onrender.com
• Frontend (Streamlit): https://guitar-assistant.streamlit.app/

Repository

• GitHub: https://github.com/EbenTheGreat/research-assistant

Setup (Local Development)

git clone https://github.com/EbenTheGreat/research-assistant
cd research-assistant
pip install -r requirements.txt

# Run backend
uvicorn app.main:app --reload

# Run frontend
streamlit run streamlit_app.py

Environment variables:

• VOYAGEAI_API_KEY
• PINECONE_API_KEY
• GOOGLE_VISION_KEY
• GROQ_API_KEY

Model Selection Rationale

Each model and service used in PedalBot was selected to balance accuracy, performance, and efficiency within the RAG pipeline:

• OCR – Google Vision API:
  Chosen for its reliability in extracting text from both digital and scanned PDFs. It handles varied fonts and layouts effectively, which is crucial for processing complex pedal manuals.

• Embeddings – VoyageAI (voyage-3.5):
  Selected for its strong semantic understanding and optimized cost structure. The voyage-3.5 text embedding model provides high-quality contextual representations while minimizing API costs and rate-limit constraints—making it well-suited for scalable retrieval tasks.

• Vector Database – Pinecone:
  Provides efficient and low-latency vector search for large embedding collections. Pinecone’s managed infrastructure ensures fast and accurate retrieval across multiple manuals without manual index tuning.

• LLM – Groq (llama-3.1-8b-instant):
  Used for its blend of reasoning quality and inference speed. Groq’s hosted LLaMA model supports near real-time response generation while maintaining cost efficiency, ideal for interactive assistant scenarios.

Screenshots

Frontend Query Example

Upload Manuals Interface

Future Improvements

• Expand supported document formats (HTML, TXT).
• Improve multi-page citation formatting.
• Optimize retrieval for larger knowledge bases.

Summary

PedalBot is a foundational RAG assistant that enables guitarists to instantly query effect pedal manuals with source-backed responses. By combining embeddings, vector search, and LLM generation, it delivers practical AI assistance with transparency and accuracy.