Truth Seeker

• 

  Example 1: Single-sentence Claim

• 

  Example 2: Paragraph Claims

• 

  Example 3: Document Claims

Inspiration

I was inspired to build Truth Seeker by the overwhelming amount of pseudoscience and health myths circulating in social media and family group chats. I wanted to create a tool that doesn't just "chat," but actually investigates claims like a researcher to help people make informed decisions.

How Gemini 3 Powers the App

Gemini 3 Pro is the absolute core of this application, utilized in three critical ways:

1. Multimodal Analysis: We use Gemini’s vision capabilities to let users upload images of supplement labels or clickbait article screenshots, automatically extracting the central claim for verification.
2. Function Calling (Tool Use): This is central to the architecture. Gemini acts as an autonomous agent that translates natural language doubts into precise, boolean search queries for the Semantic Scholar API.
3. Complex Reasoning: Instead of hallucinating answers, Gemini processes the raw JSON data from academic abstracts. It synthesizes conflicting evidence to generate a nuanced "Verdict" (Scientific Consensus, Debated, or Pseudoscientific).

How We Built It & Challenges

Built with Next.js 15 and FastAPI, the biggest challenge was coordinating the strict rate limits (1 RPS) of the academic database with the user's need for speed. We solved this by using Python's asyncio to queue requests while Gemini managed the conversation flow. Through this project, I learned that the true power of LLMs lies not just in generation, but in their ability to act as a reasoning engine for external, authoritative data (RAG).

Github link: https://github.com/feixukeji/Truth-Seeker

Built With

• asyncio
• docker
• fastapi
• google-cloud-run
• google-genai
• next.js
• python
• react-markdown
• semantic-scholar-api
• tailwind
• typescript