This is a submission for the . Below are some screenshots demonstrating the functionality of the app:
Tools Used
Timescale Cloud: Database hosting
pgVector: For efficient data storage and retrieval of embeddings
pgai Vectorizer: For enhanced text processing and chunking within the database
Ollama: For generating embeddings and handling response generation
Streamlit: Used for building the user interface
Final Thoughts
Building this application has been an insightful journey, especially when it comes to understanding how vector databases work and experimenting with various indexing strategies. One of the key concepts I explored was HNSW (Hierarchical Navigable Small World), a graph-based algorithm that significantly improves search retrieval performance. Although I didn't implement HNSW in this initial version due to the relatively small dataset, it’s something I plan to explore further in the future.
I also dived deep into pgai, a suite of tools designed to simplify the development of RAG (Retrieval-Augmented Generation) systems and semantic search applications with PostgreSQL. pgai has made it easier to manage and query embeddings directly within the database, which simplified many aspects of this project.
Additionally, I go through the source code for pgai and gained a better understanding How pgai use sql extension feature to interact with LLM.
A funny thing is that I haven’t been able to fully test OpenAI Mode yet, because I don’t have a paid OpenAI account. So, if you encounter any issues or bugs, feel free to reach out to me—I’d be happy to assist!
Prize Categories:
Open-source Models from Ollama, Vectorizer Vibe, All the Extensions
Future Improvements
While time constraints limited what I could implement for this project, I have plans to add the following features in the future:
- Pre-Retrieval Optimization using Query Transformation techniques
- Implementing Late Chunking
- Evaluating Chunking Strategies using Recall, with LLMs as judges
- Retrieval Optimization using Ensemble/Fusion Retrievers
Finally, all the best to everyone participating in the hackathon!
SOCIAL SHARE CARD GENERATOR