ZhiLog

An AI learning agent that learns with you — not just answers you.

Comment

Inspiration

As graduate students, we constantly faced fragmented learning workflows — switching between ChatGPT, PDF viewers, note-taking apps, and cloud drives. We wanted a better way to learn deeply, not just quickly. That’s why we built ZhiLog: an AI-native learning agent that helps students complete the “read → ask → note → retain” cycle seamlessly.

What it does

ZhiLog allows users to:

  • Upload PDFs or insert article links
  • Ask contextual questions and get source-backed answers (via Perplexity Sonar)
  • Auto-generate structured notes in Markdown
  • Build a personal knowledge graph from interactions
  • View and reuse their Q&A and notes in future sessions

How we built it

  • Backend: Python, LangChain, Perplexity Sonar API
  • Frontend: Flask(demo), Figma (UI design)
  • Parsing: PyPDF2, BeautifulSoup

Challenges we ran into

  • Maintaining context in multi-page document conversations
  • Designing a simple but powerful UX that balances control and automation
  • Managing LLM cost vs. accuracy during API integration
  • Building a scalable knowledge graph backend within time limits

Accomplishments that we're proud of

  • Built a functional prototype with full “read–ask–note” flow
  • Integrated Perplexity API for trustworthy, real-time answers
  • Designed an interface learners said they would use daily

What we learned

  • Perplexity's Sonar API is ideal for fast and factual Q&A
  • Learners value context and memory over flashy features
  • Agent-native workflows reduce cognitive load and boost retention
  • Great UX means one clear entry point, not a dozen scattered tools

What's next for ZhiLog

  • Expand knowledge graph capabilities
  • Integrate Google Drive and Canvas for dynamic material fetching
  • Launch ZhiHub: a community for shared knowledge bundles
  • Agent operation visualization: boost trust among users
  • Continue iterating based on user testing and agent behavior feedback
  • Prepare for MVP launch and seed round fundraising

Built With

  • beautifulsoup4
  • duckduckgo-search-api
  • flask
  • pypdf2
  • python
  • werkzeug
Share this project:

Updates