# Project export: SocratesAI This document was generated by HackStack to give an AI agent context about a hackathon project. Sections are labeled with their provenance; content marked as truncated was cut to keep this document small. ## Project metadata - Hackathon: TreeHacks 2025 - Tagline: SocratesAI is an AI agent that monitors your Zoom Calls, fact-checking in real-time every statement, ensuring reliability and trustworthiness of the discussion while enriching it with useful insights. - Devpost: https://devpost.com/software/socrates-ai - GitHub: https://github.com/AshwathKarunakaram/treehacks25 - Video: https://www.youtube.com/embed/UUcRo7pdvPM?enablejsapi=1&hl=en_US&rel=0&start=&version=3&wmode=transparent - Team: 2 GitHub contributor(s) — MatteoLeonforte (1 commits), AshwathKarunakaram (1 commits) ## Devpost submission (written by the team) ### Inspiration In today’s digital world, misinformation spreads rapidly, influencing critical areas like politics and health. With the exponential growth of online information —both true and misleading— fact-checking has never been more important. AI has revolutionized research, enabling in-depth analysis at a speed and depth impossible for humans. We believe AI should play a crucial role in ensuring discussions remain informed, reliable, and insightful. Inspired by Socrates' pursuit of truth, we built SocratesAI, an AI-powered real-time fact-checking assistant for Zoom calls. What It Does SocratesAI is an AI agent that actively listens to conversations during Zoom meetings, continuously fact-checking statements by searching the web in real time. It not only verifies the reliability of spoken information but also enriches discussions with relevant insights and supporting data. At the end of each call, it provides a summary report with key insights, flagged inaccuracies, and recommendations for deeper exploration. How We Built It SocratesAI combines multiple cutting-edge technologies to deliver real-time fact-checking: Zoom Real-Time Media Stream (RTMS): Allows us to access the conversation’s transcript using Zoom’s built-in voice-to-text converter. A contextual sliding window dynamically updates with the conversation’s key points using OpenAI ChatGPT-4o, ensuring long-term context retention without excessive input size. Perplexity Sonar then evaluates the last conversation chunk and context, determining truthfulness, providing explanations, and linking to reliable sources. To enhance engagement, the response is converted into speech using ElevenLabs' text-to-speech model Flash V2, allowing the AI to deliver information smoothly within the conversation. Challenges We Ran Into Real-time processing: we needed to balance response speed and accuracy. We solved this by batching the transcript into small but meaningful segments that preserve context while keeping latency low. Understanding irony and figurative speech: careful prompting helped improve AI comprehension and minimize misinterpretations. Minimizing disruption: instead of interrupting speakers, SocratesAI provides subtle auditory cues and a visual popup with a brief fact-check summary. Users can choose to ignore the cues or let the AI deliver a full audio response. Accomplishments That We're Proud Of Seamlessly integrating multiple AI models and services into a single, real-time fact-checking system orchestrated with LangChain. Designing an AI-human interaction model that enhances conversations without being intrusive. What We Learned User experience is critical—our peer surveys helped refine interaction preferences and ensure minimal disruption. Latency vs. accuracy trade-offs—finding the optimal balance between quick responses and detailed context analysis is key. The incredible impact on society of misinformation and the urgency for robust solutions to check information's reliability. What's Next for SocratesAI Enhancing answer quality and reducing latency. Improving the AI’s understanding of irony, sarcasm, and figurative speech. Expanding to multilingual interactions. Adapting SocratesAI for critical environments such as classrooms, legal courts, and professional debates where information accuracy is essential. ## README (from the GitHub repository) # treehacks25 ## Zoom mock server instructions Documentation here: https://github.com/ojusave/mockRTMSserver ### Setup 1) Install Docker 2) Clone repo (link) 3) Build container (see intstruction at link) ### Mock Server 1) Run container 2) Open address (server): http://localhost:9092 ### Client 1) Run 'mockRTMSserver/node test_client/server.js' 2) Run 'ngrok http 8000' 3) Copy address URL into the Server at http://localhost:9092 (should receive confirmation on client) 4) Start recording on the server and receive datastream on the client ### TODO Modify (a copy of) test_client/server.js In lines 181-207 there is the function executed any time a new message is received from the Zoom mock server. This stream can be forwarded to another websocket (atm at the address "http://localhost:5000/process_media"), that can be read by the LangChain Server (not sure how that part works). ## Detected evidence (automated analysis) Indexed codebase: 1 recognized source files, 1 KB. - AWS (technology) — claimed on Devpost, not found in the code - FastAPI (technology) — claimed on Devpost, not found in the code - JavaScript (language) — claimed on Devpost, not found in the code - LangChain (technology) — claimed on Devpost, not found in the code - Next.js (technology) — claimed on Devpost, not found in the code - OpenAI (technology) — claimed on Devpost, not found in the code - Python (language) — claimed on Devpost, not found in the code - Tailwind CSS (technology) — claimed on Devpost, not found in the code ## Codebase structure (from repository index) ### Files (1 of 1) ``` README.md ``` ### Dependencies No dependency index available. ### Recent commits (newest first) - Instructions for Zoom data stream reading - Initial commit ## Key source files (fetched from GitHub, selected and truncated for size) No suitable source files found.