FocusCam

Staying focused is hard. FocusCam leverages AWS Rekognition to detect distraction through facial cues and visualize focus patterns to help students stay on track.

Comment

Inspiration

We wanted a lightweight, privacy-aware way to understand when someone is focused during a session (study, meetings, deep work) using only the browser and a serverless backend—no heavy installs or complex setup.

What it does

FocusCam captures webcam frames in the browser and periodically sends them to an AWS Lambda via API Gateway. The Lambda uses Amazon Rekognition signals to estimate a simple "focus" score and records events in DynamoDB. An analytics page can visualize trends over time.

How we built it

  • Frontend: a static site (frontend/index.html, script.js, style.css) that requests webcam access, samples frames every ~5s, and POSTs JSON with sessionId, timestamp, and imageBase64 to an API endpoint.
  • Backend: a Python Lambda (lambda/processFocusImage.py) that calls Rekognition, computes a focus score, and writes results to DynamoDB. Exposed via API Gateway.
  • Dev loop: run the frontend locally with a static server; test the Lambda handler locally with a small Python snippet and boto3.

Challenges we ran into

  • Handling browser permission flows reliably across different devices
  • Designing a simple, explainable focus metric from Rekognition outputs
  • Ensuring secure, authenticated access before any public exposure
  • Ensuring our API Gateway and backend are configured to correctly handle CORS requests

Accomplishments that we're proud of

  • End-to-end working prototype with minimal infra
  • Clean separation between a simple frontend and serverless backend
  • Straightforward payload and handler that are easy to test locally

What we learned

  • Practical tradeoffs of browser-based frame capture and rate limiting
  • Rekognition signal quality varies with lighting, angle, and camera
  • DynamoDB works well for append-only event logging and quick analytics

What's next for FocusCam

  • Add authentication (Cognito/API keys/IAM) and per-user sessions
  • Improve focus scoring heuristics and/or add light on-device ML
  • Integrate Amazon Bedrock to generate personalized tips to improve focus scores
  • Add unit tests for the Lambda (moto/pytest) and a minimal CI
  • Provide infrastructure-as-code (SAM/CloudFormation) for one-command deploy
  • Ship a simple hosted analytics dashboard

Built With

Share this project:

Updates