Gspace Chat

A retrieval-augmented AI assistant that turns Gilmour Space’s public website into a searchable knowledge interface with grounded answers and source citations.

1. Project overview

Gspace Chat is a Python Streamlit application that answers user questions using selected public pages from the Gilmour Space website. The application uses a local retrieval pipeline: it scrapes website content, cleans and chunks the text, builds a local ChromaDB vector index, retrieves relevant chunks for each question, and generates answers grounded only in retrieved content.

2. Why this project exists

Public website information is often spread across multiple pages and can be time-consuming to search manually. This project provides a focused, query-based interface over that website content, while keeping answers tied to source material and reducing unsupported responses.

3. Current MVP features

• Streamlit-based chat interface
• Scraping of selected public Gilmour Space website pages
• Text cleaning for scraped website content
• Chunking pipeline for retrieval-ready documents
• Local ChromaDB vector index generation
• Retrieval of relevant chunks at question time
• OpenAI chat model response generation grounded in retrieved content
• Source links shown under answers
• Fallback response when evidence is missing: I don’t know from the Gilmour Space website.
• Streamlit sidebar admin control to rebuild the local index

4. Architecture

Gilmour Space website pages
        ↓
Scraper
        ↓
Cleaner
        ↓
Chunker
        ↓
ChromaDB vector index
        ↓
Retriever
        ↓
OpenAI grounded answer generation
        ↓
Streamlit UI with citations

5. Tech stack

• Python
• Streamlit
• ChromaDB
• OpenAI API (chat + embeddings)

6. Setup

Windows PowerShell:

python -m venv .venv
.\.venv\Scripts\activate
pip install -r requirements.txt
Copy-Item .env.example .env

7. Environment variables

Set these values in .env:

OPENAI_API_KEY=
OPENAI_CHAT_MODEL=gpt-4.1-mini
OPENAI_EMBEDDING_MODEL=text-embedding-3-small

8. Build the local knowledge base

Run the pipeline steps in order:

python -m gspace_chat.scraper
python -m gspace_chat.cleaner
python -m gspace_chat.chunker
python -m gspace_chat.indexer

You can also rebuild the local index from the Streamlit sidebar admin button.

9. Run the app

streamlit run app.py

10. Example questions

• What is Gilmour Space’s Eris rocket?
• What launch services are described on the website?
• Where is Gilmour Space’s launch site located?
• What does the website say about current mission or launch updates?

11. Hallucination control

The assistant is designed to answer from retrieved website chunks only. If relevant evidence is not found in retrieved content, it returns:

I don’t know from the Gilmour Space website.

This behavior is intended to reduce unsupported claims.

12. Repository safety / secrets

• Do not commit .env.
• .env, data/, chroma_db/, __pycache__/, and .venv are gitignored.
• data/ and chroma_db/ are local generated folders and are not committed.

13. Known limitations

• This is an MVP, not a production system.
• Retrieval quality depends on page selection and chunking strategy.
• Answers are limited to scraped website content and may miss information not present in indexed pages.
• Content freshness depends on when the scraping and indexing pipeline was last run.

14. Future improvements

• Expand and manage source page coverage.
• Improve chunking and retrieval tuning for better precision and recall.
• Add stronger admin controls for index lifecycle and content refresh.
• Add evaluation workflows for answer quality and citation quality.
• Add production-oriented concerns such as observability, authentication, and deployment hardening.