[WIP] Debug the project and improve mudra detection

[Copilot]

• Fix frontend/.env.local port mismatch (5001 → 5000)
• Fix mudraDetector.js Katakamukha detection bug (ring finger check inverted)
• Improve mudraDetector.js: normalize distances by hand span, improve Mayura/Pataka/Tripataka logic, add temporal smoothing
• Fix Practice.js: remove broken script-tag injections, add MediaPipe load-wait polling, integrate temporal smoothing, fix per-frame stats inflation, improve camera permission error handling
• Validate build and run tests

Original prompt

Repository: hazelr1/nrityakala

Goal: Debug the entire project and make it fully functioning, with emphasis on correct mudra detection.

Please perform a complete end-to-end stabilization and improvement pass:

1. Reproduce and diagnose

• Clone and run the project locally using documented or inferred commands.
• Identify all runtime/build/lint/test errors that prevent normal operation.
• Document root causes clearly.

2. Fix core project health

• Resolve dependency/version issues and broken imports.
• Fix configuration or environment issues needed for successful startup/build.
• Ensure the app starts and runs without crashes.
• Add or update scripts in package.json if missing for reliable dev/build/test/start workflows.

3. Mudra detection correctness

• Trace the current mudra detection pipeline (camera input, landmark extraction, gesture classification logic, thresholds, smoothing/debouncing).
• Fix logic bugs that cause false positives/false negatives.
• Improve robustness for practical usage (e.g., handedness handling, mirrored webcam assumptions, normalization/scaling invariance, temporal smoothing).
• If model/config files are missing or miswired, repair wiring and loading paths.
• Ensure detection results are mapped to the intended mudra labels correctly.

4. UX and reliability improvements needed for functionality

• Add clear user feedback for camera permissions, loading states, and detection confidence/status.
• Prevent common failure states (null camera stream, model load race conditions, unhandled promise errors).
• Keep changes minimal but sufficient to deliver a fully functioning app.

5. Validation

• Verify the project runs successfully after changes.
• Validate that mudra detection works correctly for supported mudras, and reduce flicker/misclassification.
• Add/update lightweight tests where feasible for detection utility logic.

6. Delivery

• Create a PR with cohesive commits.
• Include a concise PR description summarizing:
  • what was broken,
  • what was fixed,
  • how mudra detection was improved,
  • how to run and verify.

Constraints:

• Preserve existing project intent and architecture as much as possible.
• Do not perform unrelated refactors.
• Prefer deterministic logic improvements over ad-hoc hardcoding.
• Keep dependency additions minimal and justified.

The following is the prior conversation context from the user's chat exploration (may be truncated):

User: debug the entire project, and make it a fully functioning project

it should detect the mudras correctly

Assistant: I can do that, but I need one specific detail before I can start: do you want me to open a Pull Request with the fixes in hazelr1/nrityakala?

If yes, I’ll proceed end-to-end (diagnose issues, implement fixes, and tune mudra detection accuracy).
If you want, I can also target a specific branch—otherwise I’ll use the repo default branch.

User: yes

Assistant: Thanks — I’m waiting on your confirmation in the GitHub approval dialog to proceed with the full debugging and PR creation.

This pull request was created from Copilot chat.