Turn a folder of documents into a clickable graph of people, places and claims
Advanced28h2–4
LLMPythonECharts
15/20AI Answer Claim Verifier
Paste an AI answer and see which claims are supported by real sources.
Advanced26h2–3 people16/20 fit
Build AI Answer Claim Verifier for students and researchers using chatbots in ~26h with 2–3 people. Stack: LLM + Search API + RAG + Next.js.
Why it wins: Directly attacks hallucinations, which every judge worries about.
Build flow
01 · Problem
Chatbots state wrong facts confidently and users cannot tell which parts to trust.
02 · Solution
Split the answer into atomic claims, search for sources, check each claim against retrieved passages and colour it supported, unsupported or unclear.
03 · Build
- 1Prompt to split claims
- 2Search and fetch passages per claim
- 3Ask the LLM for a verdict with a quote
- 4Colour-code the answer
04 · Demo
Paste a chatbot answer with one wrong date and watch it turn red with a source.
MVP checklist
- Claim splitting
- Per-claim source search
- Support verdicts
- Source links
Tech stack
| Tech | Role | |
|---|---|---|
| LLM | Language model | AI |
| Search API | Web search | API |
| RAG | Retrieval over your docs | AI |
| Next.js | Frontend + API routes | Web |
Related ideas
More AILets people ask questions about long documents instantly
Intermediate24h1–3Next.jsRAGLLM16/20See how different outlets frame the same news story side by side
Intermediate16h1–3Search APIEmbeddingsLLM16/20