CampusLens AI: A Gemma 4 Multimodal Assistant for College Notices

College students receive important information through notices, posters, assignment sheets, and other visual documents. Finding the important details manually can be time-consuming and confusing.

For the MLH Hacktoberfest Hack Day Coimbatore, our team Synap Tech built CampusLens AI, a multimodal AI assistant that uses Gemma 4 to understand college notices, extract important information, and answer questions about the document.

What We Built

CampusLens AI allows a student to upload a photo or image of a college notice.

The system then:

1. Understands the uploaded document using Gemma 4.
2. Extracts important information such as:
   • Event
   • Date
   • Time
   • Venue
   • Deadline
   • Requirements
3. Generates a concise summary.
4. Allows the student to ask questions about the uploaded notice.

For example, instead of manually reading a long event poster, a student can ask:

"What do I need to bring?"

or

"When is the event?"

and CampusLens AI provides an answer based on the uploaded document.

How We Built It

The main technologies used in our project are:

• Gemma 4 for multimodal document understanding and question answering
• Google GenAI SDK for interacting with Gemma 4
• Python for the AI processing
• Streamlit for the web interface
• Git and GitHub for collaborative development

Our core workflow is:

Upload Document → Gemma 4 → Understand → Extract Information → Ask Questions

We used prompt engineering to make the model return structured information and instructed it not to guess information that is not present in the document.

Why Gemma 4?

Gemma 4 was a great fit for this project because CampusLens needs to work with visual documents rather than only plain text.

The multimodal capabilities allow the model to understand information directly from an uploaded notice and combine that understanding with natural-language question answering.

Challenges We Faced

During development, we encountered a few practical challenges.

One challenge was model availability and stability. We initially tested gemma-4-31b-it, but experienced server errors on one of our development machines. We tested the available Gemma 4 models and switched our working application to gemma-4-26b-a4b-it, which worked reliably for our multimodal workflow.

We also had to debug API connectivity, image processing, Streamlit integration, and Git collaboration issues while working as a team.

Because the hackathon had a limited development window, we focused on making the core workflow reliable rather than adding too many extra features.

What We Learned

Through this project, we learned about:

• Multimodal AI and document understanding
• Gemma 4 integration
• Prompt engineering
• Building AI-powered interfaces with Streamlit
• API integration and debugging
• Collaborative Git and GitHub workflows
• Rapid MVP development during a hackathon

Future Scope

There are several features we would like to explore in the future:

• Calendar integration for extracted events
• Smart reminders for deadlines
• Action-item extraction
• Support for multiple document types
• Multilingual document understanding
• Searching across multiple uploaded notices
• Mobile-friendly improvements

Team

Synap Tech

• Lakshithaa V — AI/backend development, Gemma 4 integration, multimodal document processing, information extraction, Q&A, Git/GitHub
• Pavitraa Surendran — UI and frontend development using Streamlit
• Shivamihit G — Documentation and presentation
• Rithvik Kumar — Prompt engineering, AI testing, and edge cases

Links

GitHub Repository

https://github.com/lakshiscooby-alt/campuslens-ai

Demo Video

https://youtu.be/EYpPQo6tEKM?si=82X_DOr5VlY8jvKz

Built For

This project was built during MLH Hacktoberfest Hack Day Coimbatore.

We are particularly submitting CampusLens AI for the Best Use of Gemma 4 challenge.

Thanks to the organizers and mentors for giving us the opportunity to build and experiment with Gemma 4!