https://github.com/epsilla-cloud/docs
Epsilla Documentations
https://github.com/epsilla-cloud/docs
Last synced: 5 months ago
JSON representation
Epsilla Documentations
- Host: GitHub
- URL: https://github.com/epsilla-cloud/docs
- Owner: epsilla-cloud
- License: apache-2.0
- Created: 2023-07-28T05:49:17.000Z (almost 3 years ago)
- Default Branch: main
- Last Pushed: 2025-03-05T14:09:57.000Z (over 1 year ago)
- Last Synced: 2025-03-05T15:20:08.504Z (over 1 year ago)
- Size: 71.3 MB
- Stars: 0
- Watchers: 1
- Forks: 1
- Open Issues: 1
-
Metadata Files:
- Readme: README (1).md
- License: LICENSE
Awesome Lists containing this project
README
# Overview
Epsilla offers an open-source vector database. Our focus is on ensuring scalability, high performance, and cost-effectiveness of vector search. Epsilla bridges the gap between information retrieval and memory retention in Large Language Models. We see ourselves as the **Hippocampus for AI**.
## Common use cases
Here are some common use cases of Epsilla vector database
### 1. Augmenting LLMs with Proprietary Data
**Problem:** LLMs don’t have latest knowledge about the world (e.g., GPT-4 has a knowledge cutoff of April 2023), and don’t have knowledge about any private data (e.g., your company's knowledge base)
**Our solution:** Augment LLMs by adding semantically similar information retrieved from vector database into the prompt (also known as [Retrieval Augmented Generation](https://ai.meta.com/blog/retrieval-augmented-generation-streamlining-the-creation-of-intelligent-natural-language-processing-models/)).
**Benefits:** Enable the LLMs to work for your own data and knowledge. Compare to using fine-tuning, RAG has a much faster time-to-value, is much cheaper for both engineering cost and hardware cost, and support real time knowledge updates.\
.png)
Example: Upload IRS tax publications
Example: Build a tax assistant chatbot augmented by IRS publications
### 2. Build Better Recommendation Systems
**Problem:** It’s really hard to improve recommendation result relevance, and even harder to build a scalable realtime recommendation system.
**Our Solution:** Use embedding as the bridge between incomparable data types to leverage the hidden relevance of user behavioral data during recommendation.
**Benefits:** Vector DB that leverages the hidden relevance improves recommendation recall. Epsilla’s low query latency is vital to building a realtime recommendation system.
### 3. Find Hidden Insights From Unstructured Data
**Problem:** It’s really hard to analyze and query unstructured data (images, audios, videos) based on their content semantics.
**Our Solution:** Connect and index the unstructured data based on the semantic relevance of their content, and enable multimodal search and analytics.
**Benefits:** Multimodal search becomes as easy as text search. No need to manually label unstructured data and convert to structured data anymore.
.png)
