HackFest GDG Loyola 2026 · Ateneo De Manila University · March 2026
VerifAI: AI-Powered Source Verification for Student Research Papers
Lead Developer · Backend & AI Integration
At HackFest GDG Loyola, my team built VerifAI, a research assistant prototype set inside a Google Docs style editor. A student highlights a claim in their paper, and the tool finds related peer-reviewed literature, checks that each source exists, and shows where the paper supports or contradicts the claim.
I led development of the backend and AI integration, covering the Flask server, the Gemini and CrossRef pipeline, and the credibility rubric. My teammates focused on the interface and visual design.
- Built the Flask backend and a server-side proxy for the Gemini API, so the API key never reaches the browser
- Designed a two-step source check: Gemini suggests papers, then parallel CrossRef lookups confirm each DOI, title, and author list
- Developed the 100-point credibility rubric, which scores each source on recency and relevance
- Wrote the technical documentation, including data flow diagrams, use cases, and the SRS
How it works
1. Discover
Gemini returns candidate papers for the highlighted claim. The backend verifies each one against CrossRef and falls back to a Google Scholar search when no match is found.
2. Ground
The student pastes the paper's text. Gemini labels the result as an exact match, a related passage, or a contradiction of the claim.
3. Review
Relevant passages are highlighted in the paper by color, alongside the rubric verdict, an APA citation, and a synthesis paragraph.
- Python
- Flask
- Gemini 2.5 Flash
- CrossRef API
- JavaScript
- HTML/CSS
The main lesson I took from this project was how to make LLM output dependable: rather than trusting the model's answers directly, I validated them against an authoritative source before showing them to the user.
