Inspiration
What it does
KnoWhere is harnessing language models for creating fast, efficient and accurate insights over customer databases.
How we built it
We created Webservices and a Cross-Platform App to enable the customers for their mission. We used Dart, Python in combination with elastic search and the PALM LLM for a good solution.
Challenges we ran into# Sika Challenge - HackZurich 2023
Prompt Engineering and good UI Experience.
Project Overview
The "Sika Challenge" project was developed during HackZurich 2023 with the aim of addressing the challenges associated with a vast, unstructured, and isolated knowledge base. These challenges include difficulties in finding relevant sources and limited real-time access to knowledge.
Problem Statement
In today's information age, access to knowledge is essential for decision-making and problem-solving. However, many knowledge bases are unstructured, making it challenging to find relevant information quickly. Additionally, real-time access to knowledge can be a significant bottleneck in critical situations.
Solution
The "Sika Challenge" project offers a solution to these issues by harnessing the capabilities of Large Language Models and Word Embeddings. We have developed a context-driven search system that empowers users to ask questions freely. The system then provides summarized answers based on the knowledge base, complete with references to the original sources.
Key Features
Context-Driven Searches: Our system uses advanced language models and word embeddings to understand and interpret user queries in context.
Summarized Answers: Users receive concise and relevant answers to their questions, reducing the need to sift through extensive documents.
Source Attachments: Each answer is accompanied by references to the sources, enabling users to verify and explore further if needed.
File Upload: New knowledge can easily be added to the knowledge base. Just upload the according pdf file on the website and it can immediately be found using the search tool.
How It Works
Ask Your Question: Users can input their questions or queries into the system.
Contextual Analysis: The system employs language models and word embeddings to understand the context and nuances of the question.
Knowledge Retrieval: Relevant information from the vast knowledge base is retrieved based on the query.
Summarized Response: Users receive a concise and informative answer that addresses their query.
Source References: To ensure credibility, the response includes links or citations to the original sources.
Getting Started
To get started with the "Sika Challenge" project, follow these steps:
- Visit http://knowhere.westeurope.cloudapp.azure.com:8080/
- Question Statement: Write an arbitrary question into the given window. It can be keywords, a problem description or a natural language question.
- Knowledge Overview: The system proposes the best fitting sources to you and enables you to easily fact check them, sorted by relevance.
- Content Summary: Our Knowledge Hub will summarize the most important knowledge within each paper regarding your question, and provide you with a concise answer.
Accomplishments that we're proud of
As a team we built all together a working prototype from scratch. And we are proud of what we accomplished together and individually technically and also on a human level.
What we learned
Information retrieval, Prompt Engineering, Web services, front end development, team spirit, presentation and pitching.
What's next for KnoWhere
If there is enough interest, we would like to work on on it and create potentially our own start-up.
Log in or sign up for Devpost to join the conversation.