README.md

metadata

title: TruthLens
emoji: 🦀
colorFrom: red
colorTo: pink
sdk: gradio
sdk_version: 5.46.0
app_file: app.py
pinned: false
license: apache-2.0
short_description: 'Here’s TruthLens – Lite: a public demo that verifies claims '

🧭 TruthLens – Misinformation-Aware RAG

TruthLens is a fact-checking AI demo that shows how modern GenAI can be made responsible, transparent, and safe.
It verifies controversial claims (vaccines, climate, elections, etc.) using retrieval-augmented generation, checks its own answers for faithfulness, and applies safety filters.

---

🚀 How it works

1. Input a claim/question
   Example: “Did humans cause global warming?”

2. Retrieve sources
   The app looks at trusted sources (WHO, IPCC, election reports). You can also paste your own text.

3. Generate grounded answer
   TruthLens writes an answer with [1], [2] style citations from those sources.

4. Self-check
   It tests how well the answer actually matches the sources (faithfulness score).

5. Safety filters

   • Removes personal info (PII redaction).
   • Flags toxicity levels if the language is harmful.

---

🛠️ Tech stack

• Gradio (UI & deployment)
• Transformers (Flan-T5 generation, DeBERTa NLI, BERT NER, Toxic-BERT)
• SentenceTransformers (MiniLM embeddings)
• CrossEncoder (MS-MARCO reranker)
• scikit-learn / pandas / numpy / matplotlib

All models are CPU-friendly, so the demo runs in a free Hugging Face Space.

---

🌍 Why it matters

• Tackles misinformation around health, climate, and politics.
• Demonstrates responsible AI practices (grounding, self-checking, safety).
• Shows how a Lead AI Developer designs not just models, but systems ready for production.

---

▶️ Try it

1. Enter a claim or question.
2. (Optional) Paste your own text sources.
3. Click Run TruthLens → see answer, citations, faithfulness score, and safe-share version.

---

👔 About this project

Created by Aso Bozorgpanah as a portfolio demo for Lead AI Developer roles.
Focus: Explainability, safety, and production-readiness in Generative AI.