PerspicuaAI

PerspicuaAI is a decentralized platform designed to revolutionize research by tackling information overload. Our AI Agent uses advanced natural language processing to generate concise summaries of research papers from diverse sources, going beyond simple abstract condensation to extract key findings and methodologies. This intelligent summarization empowers researchers to quickly grasp the essence of complex studies, saving valuable time and accelerating knowledge discovery. PerspicuaAI's architecture is built for adaptability, allowing it to integrate with various research data sources and evolve with advancements in AI.

A core principle of PerspicuaAI is trust and transparency. AI-generated summaries undergo rigorous decentralized verification using EigenLayer's Predicate AVS. A network of validators independently assesses summaries, running our custom verification logic to ensure accuracy and quality. This process, including semantic similarity checks and other criteria, is recorded on-chain, providing verifiable proof of the summary's trustworthiness. This decentralized verification mechanism builds confidence in the information presented, mitigating the risk of bias or inaccuracies.

Verified summaries, along with original papers and metadata, are immutably stored on IPFS, a decentralized storage network. This ensures open, persistent, and censorship-resistant access to research insights, democratizing knowledge and fostering global collaboration.

PerspicuaAI is built using a combination of cutting-edge technologies, seamlessly integrated to deliver efficient, trustworthy, and accessible research insights. Our implementation leverages Python as the primary programming language, chosen for its rich ecosystem of scientific computing and AI libraries.

AI Summarization Engine (AI Agent): At the heart of PerspicuaAI lies our sophisticated AI Agent, responsible for generating concise and informative summaries. This agent is implemented as a modular Python class, allowing for easy expansion and integration of new summarization techniques. Currently, the AI Agent utilizes the spaCy library for natural language processing (NLP) tasks, including tokenization, part-of-speech tagging, and named entity recognition. Future iterations will explore more advanced abstractive summarization models, potentially leveraging transformer networks.

Decentralized Verification (EigenLayer Predicate AVS): A key innovation of PerspicuaAI is the integration of EigenLayer's Predicate AVS for decentralized verification. The AI Agent is responsible for interacting with the Predicate AVS. When a summary is generated, the AI Agent formats the summary according to the AVS API specifications and sends a verification request to the AVS endpoint. We use the requests library in Python to handle the HTTP communication with the AVS. The AVS, in our current setup, is a simulated predicate that verifies the length of the summary. However, this will be replaced with a more sophisticated predicate as we develop the project further. The response from the AVS, containing the verification status (True/False) and a message, is then processed by the AI Agent.

Immutable Storage (IPFS Integration):- To ensure open access and censorship resistance, verified summaries, along with the original abstracts and metadata (title, link, DOI), are stored on the Interplanetary File System (IPFS). The AI Agent, after receiving a successful verification from the Predicate AVS, packages all relevant data into a JSON object. We use the ipfshttpclient library in Python to interact with the IPFS API. We use the /api/v0/add?wrap=true endpoint to add the JSON data to IPFS, which creates a directory structure containing the data file and returns the root CID (Content Identifier) of the directory. This ensures that the data is wrapped in a directory, which is important for retrieving the complete JSON object later. This CID is then stored and used for retrieval.

BioRxiv API Integration:- PerspicuaAI retrieves research abstracts and metadata from BioRxiv using their public API. We utilize the endpoint, as it offers a more structured and reliable way to access preprint data than the search endpoint. The requests library is used to make HTTP requests to the BioRxiv API. The JSON response is then parsed, and the necessary information (abstract, title, DOI, etc.) is extracted. URL encoding is used on the search query to ensure proper API requests.

User Interface:- For the hackathon demonstration, we have a basic command-line interface. Future development will focus on creating a user-friendly web interface usingNext.js. This interface will allow users to easily search for preprints, view summaries, and access the original research articles on BioRxiv and the stored data on IPFS.