PeerLearn AI

AI-powered collaborative verification that teaches learners to question, investigate, and reason from evidence.

Comment

🧠 PeerLearn AI — Learn to Verify, Not Just Believe

An AI-powered collaborative verification platform that turns fact-checking into a structured learning experience.

PeerLearn AI helps learners investigate claims, evaluate evidence, challenge assumptions, collaborate with peers, and form independent conclusions — instead of just being handed a "true" or "false" verdict.

AI should not tell learners what to think. It should help them learn how to think.

💡 Inspiration

We live in an age of information abundance. Students can access articles, social posts, and AI-generated content within seconds — but more information doesn't mean better understanding.

The real skill is knowing how to ask: Where did this claim come from? Is the source reliable? What evidence supports or challenges it? What conclusion can I reasonably draw?

That question became the whole premise of PeerLearn AI: turn verification from a single fact-check action into a structured educational journey.

🔎 What it does

A learner starts with a claim and works through a 5-step Socratic Investigation:

  1. Original Source Provenance — trace the claim back to where it actually came from.
  2. Independent Corroboration — find sources that support, challenge, or add context.
  3. Contradictory Evidence & Counter-Hypotheses — actively search for opposing evidence and alternative explanations.
  4. Source Reliability Appraisal — evaluate author expertise, methodology, publication context, bias, and conflicts of interest.
  5. Learner Synthesis — the learner writes their own conclusion and confidence level. The final judgment stays with them, not the AI.

The AI Mentor guides this process Socratically — instead of saying "this claim is false," it asks the learner what the original source is, what evidence supports or contradicts the claim, and what assumptions are being made.

Evidence Ledger. Learners log each piece of evidence — source, type, context, position (supporting/refuting/background), and reasoning — building a traceable trail instead of relying on one source.

Peer Verification. Other learners independently review an investigation and rate it 🟢 Verified, 🟡 Questionable, 🔴 Unsupported, or 🔵 Needs More Evidence, each with their own reasoning.

Community Consensus. Individual peer evaluations aggregate into a transparent community view — treated as evidence of community evaluation, not absolute truth. Learners are encouraged to check sources before accepting it.

Contribution Credits. Points for investigating claims, adding evidence, completing steps, and reviewing peers — a measure of participation, not factual authority.

The full lifecycle is persistent end-to-end:

Claim → Socratic Inquiry → Evidence Ledger → Peer Review → Community Consensus → Learner Synthesis → Investigation Finalized

🏗️ How we built it

Frontend: Next.js, React, TypeScript, Tailwind CSS — the investigation workspace, evidence ledger, Socratic interface, and peer/consensus views.

Backend & Data: Supabase / PostgreSQL with Row Level Security, persistent verification records, and multi-user state management.

AI Layer: Gemini-powered AI Mentor for Socratic questioning and context-aware guidance, with resilient fallback handling so the experience degrades gracefully if the AI is unavailable.

We treated this as a persistent multi-user application, not a chatbot with a UI on top — every stage of an investigation is saved so learners can leave and come back without losing progress.

🧩 Challenges we ran into

Getting from a convincing concept to a reliable system was the real challenge: database permissions and Row Level Security, persistent state after refresh, multi-user verification, preventing duplicate peer reviews, making contribution rewards idempotent, and correctly locking in the final Step 5 conclusion.

There was also a design challenge underneath the technical one: how does AI help learners verify information without replacing their judgment? Our answer was to keep the Mentor strictly Socratic — it questions, it doesn't verdict.

🚀 Accomplishments that we're proud of

We built a complete verification workflow, not just an AI chatbot:

AI guidance → evidence → structured investigation → peer review → community consensus → learner conclusion

The Step 5 synthesis is permanently recorded, and we added real safeguards against duplicate reviews, duplicate conclusions, and repeated contribution rewards — not just a happy-path demo.

🧠 What we learned

Building a trustworthy AI learning system takes more than generating good responses. Evidence provenance, persistence, permissions, peer interaction, and state management all have to work together, or the "learning" part falls apart the moment two users touch the same claim.

The bigger lesson: AI is more valuable as a thinking partner than as an authority. The most useful response isn't the fastest answer — it's the better question.

🌱 What's next for PeerLearn AI

Richer evidence discovery, stronger source comparison, more sophisticated consensus mechanisms, learning analytics, deeper multi-user collaboration, and new AI-powered learning activities — all in service of making evidence-based reasoning a practical, everyday skill.

Don't just get the answer. Learn how to verify it.

Built With

Share this project:

Updates

Submission history