Horizon Veritas

Inspiration

In our vision of a truly inclusive internet, we're committed to ensuring that everyone, regardless of their abilities or technical knowledge, can enjoy a seamless and fast browsing experience. To bring this vision to life, we've developed a groundbreaking browser extension powered by GPT technology. This extension introduces conversational applications to the entire web, addressing the limitations we found in both traditional and GPT-based conversational inputs. Our innovation doesn't stop at robust summaries and question responses; it also empowers users to take actions on web pages, making the internet more accessible and user-friendly.

In the realm of economics, our AI tool can be seen as a catalyst for market efficiency and a driver of consumer surplus. By enhancing web accessibility, we're facilitating an expansion of the consumer base, ensuring that the digital market operates closer to its potential, optimizing the allocation of goods and services to match demand. This not only streamlines consumption patterns but also amplifies the perceived value users derive from online interactions—without necessarily increasing their expenditures. Moreover, as the tool aids in the emergence of specialized job roles, it highlights the dynamic nature of human capital and the continuous need for labor market adjustments in response to technological advancements. In essence, this innovation helps the intricate interplay between technology and economic forces, underscoring the evolution of the digital economy.

What it does

Horizon Veritas is a Proof-Of-Concept, showcasing the potential of a perfect blend of chat and browsing. Positioned at the bottom of your screen, it provides semantic search and semantic-based actions. Horizon Veritas caches web page content and their respective embeddings. Upon a user's query, it performs:

Retrieval-Augmented-Generation (RAG): Providing answers with enhanced context.
Intent Extraction: Identifying user intent through vector embedding similarity, allowing for various actions like selection, filtering, and more.

How we built it

Horizon Veritas employs a novel approach using Large Language Models (LLMs) to extract user intent, compare it against predefined actions using vector embeddings, and trigger actions (assisted by GPT Function Calling). This method offers more consistency and predictability than traditional LLM Agents like Langchain.

Technologies Used:

Frontend: Chrome Extension built with ReactJS.
Middleware: Python, WebSocket.
Backend: Large Language Models (LLMs), including ADA-002 Embeddings Model, GPT 3.5 Turbo 4K, 16K, GPT 3.5 Turbo Instruct, GPT Function Calling, and Python.
Databases: Pinecone Vector Database for efficient embedding search.

Challenges we ran into

Our key challenges included:

Extracting user intent effectively.
Designing a user-friendly conversational flow.
Implementing web sockets for seamless communication.

Accomplishments that we're proud of

We're proud of achieving the following milestones:

Introducing a novel approach to intent extraction and action triggering through a Vector Database, as opposed to traditional LLM Agents like Langchain.
Successfully developing a sleek and functional Chrome extension.
Redefining the browsing experience for a more accessible and enjoyable web.