Alethia

Billions of people rely on the internet via AI or a search to get information. With vast limitations, we need to create a tool that stops the spread of misinformation.

Comment

Inspiration

The internet is currently facing a crisis of trust. With 70% of digital content now estimated to be synthetic or misleading, the "average" user is constantly bombarded by information they aren't equipped to verify. For many, this isn't just a nuisance—it's a danger. This project was born from a personal experience where my family nearly fell for dangerous medical misinformation. I realized we don’t need more AI that just "gives answers"; we need a cognitive exoskeleton that trains us how to think critically and verify the truth for ourselves.

What it does

Alethia is a multi-modal information defense system that acts as a real-time bodyguard for your digital life.

  • Real-time Forensic Analysis: Using computer vision, Alethia "sees" your active window and instantly evaluates claims on your screen.
  • Socratic Pedagogy: Instead of a simple "True/False" verdict, Alethia engages you in a Socratic dialogue via low-latency voice, challenging your logic to build long-term critical thinking skills.
  • The Forensic Profile: It detects the likelihood of AI-generation (Synthetic vs. Organic) and provides an objective bias percentage using a custom fine-tuned model.
  • Global Truth Heatmap: A synchronized web portal (try-alethia.tech) that tracks misinformation trends and debunked claims globally in real-time.

How we built it

Alethia is built on a hybrid architecture designed for extreme privacy and industrial-grade performance:

  • The Brain: An Electron desktop client that utilizes a local AMD ROCm stack with 192GB of VRAM. By running our fine-tuned evaluator locally, we ensure zero-latency responses and total data privacy.
  • The Framework: We utilized the Palantir Foundry Data Engine (FDE) to architect our initial data pipelines, ensuring that every claim is grounded in a verifiable lineage of evidence.
  • The Interface: A Next.js web portal serves as the "Global Memory," while ElevenLabs provides the natural-sounding voice for our Socratic duels.
  • The Logic: We integrated a multi-modal vision-to-logic pipeline that cross-references on-screen text with trusted knowledge nodes.

Challenges we ran into

The biggest hurdle was the hardware integration. Setting up the AMD ROCm kernel for local inference on such a large VRAM stack required significant troubleshooting of driver compatibility and memory allocation. Additionally, creating a seamless "Live Sync" between the local Electron app and the Next.js web portal required building a custom API bridge to update the Global Heatmap without compromising user privacy. Fine-tuning the Socratic engine to ensure it asks helpful questions rather than just being argumentative was an iterative process in prompt engineering and logic mapping.

Accomplishments that we're proud of

We are incredibly proud of achieving a fully local AI pipeline. In an era of cloud-dependency, running a 192GB VRAM evaluator on local hardware is a massive technical feat for a hackathon. We also successfully moved beyond a simple "chatbot" UI to create a "Forensic Lab" aesthetic that actually feels like a professional security tool. Seeing the Socratic voice successfully guide a user to debunk a lie on their own was the ultimate "aha" moment.

What we learned

This project taught us the importance of Explainable AI (XAI). We learned that users don't trust a black box; they trust a system that "shows its work." Building with the Palantir FDE taught us the value of data lineage and how to structure chaotic internet data into a usable truth ontology. On a personal level, we learned how to balance high-level "Hard Tech" (AMD/ROCm) with "Social Impact" design.

What's next for Alethia

The goal for Alethia is to become a standard browser and OS integration. We plan to:

  1. Expand the Forensic Model: Improve deepfake video detection to complement our text and image analysis.
  2. Community Socratic Duels: Allow users to invite others into a moderated, AI-facilitated debate to resolve disagreements with data.
  3. Institutional API: Provide a lightweight version of our truth engine for schools to help teach digital literacy to students in a gamified environment.

Built With

Share this project:

Updates