Masterplan

Live, editable version: https://claude.ai/code/artifact/3fee6ebc-c901-4b9b-86e9-aa16fd7ef462 This file is a snapshot from 5 Oct 2026.

Goal for Friday

By Friday evening, the wearer looks at a printed page, presses a button on the glasses, and hears the page read aloud within a few seconds. A reading screen on the laptop shows the same text with adjustable letter spacing. That screen is the first dyslexia feature, delivered without a display in the lens.

Definition of done:

• [ ] Button press on the glasses leads to speech in the earbuds in under about 8 seconds
• [ ] Reads a normal printed book page in English and Russian (Uzbek if time allows)
• [ ] Reading screen has letter-spacing, word-spacing and size sliders, and highlights each word as it is spoken
• [ ] Runs on its own battery for at least 1 hour
• [ ] 3D-printed frame is wearable, with wires hidden
• [ ] Demo video recorded as a backup for the live demo

Timeline

Software starts today without any hardware, so late parts delay only the camera step. If a checkpoint is missed, use the matching fallback in Risks.

Monday, 5 Oct: order parts and get the software working

• [ ] Order every part on the shopping list (hardware/bom/) and confirm delivery by Tuesday
• [ ] Open the reading screen (link from deploy\get_access.ps1); the server already runs on the VM
• [ ] Install Python and Tesseract (Russian and Uzbek packs) on the laptop too, as the offline fallback
• [ ] Take 10 phone photos of book pages from about 30 cm with Take photo; copy them to tests/pages/ (see docs/server.md)
• [ ] Try Seeing AI or Lookout on the same pages to see the accuracy to aim for
• [ ] Measure an existing pair of glasses and start the frame in Fusion

Checkpoint: a phone photo sent to the laptop is read aloud, with words highlighted.

Tuesday, 6 Oct: camera working

• [ ] Flash the CameraWebServer example to check the camera
• [ ] Refocus the lens to about 30 cm
• [ ] Flash firmware/glasses_cam: button press, photo, upload
• [ ] Test on a breadboard over a phone hotspot
• [ ] Print frame v1 and check fit only

Checkpoint: pressing the button on the breadboard makes the laptop read the page.

Wednesday, 7 Oct: battery power and better accuracy

• [ ] Set the MT3608 to 5.0 V with the multimeter BEFORE connecting the board
• [ ] Wire the power chain and capacitor (docs/wiring.md)
• [ ] Run on battery for 1 hour with no resets
• [ ] Improve OCR: lighting, clean-up settings, page distance
• [ ] Print frame v2 with camera mount and battery pocket

Checkpoint: fully wireless capture works on battery.

Thursday, 8 Oct: assemble and test

• [ ] Mount everything in the frame and hide the wires
• [ ] Tune the camera angle so the page is centred
• [ ] Test 20 pages, log errors in tests/results.md, fix the worst problem
• [ ] Print final frame v3 if needed
• [ ] Freeze features at 20:00. After that, bug fixes only.

Friday, 9 Oct: demo day

• [ ] Test with a real reader, ideally someone with dyslexia, and write down their feedback
• [ ] Record the demo video into demo/
• [ ] Prepare the pitch (docs/pitch.md)
• [ ] Pack the demo kit (demo/checklist.md)

3D-printed frame

• Camera board on the outside of one arm, just behind the hinge, lens forward. Battery, TP4056 and MT3608 in the other arm so the glasses balance.
• Copy dimensions from glasses that fit the wearer (front width, nose bridge, arm length).
• Camera mount angled 20-30 degrees down. Print swappable wedges at 15, 25 and 35 degrees.
• TP4056 Type-C port at the back end of the battery arm for charging.
• Wire channel along the top of the front bar (2 wires: 5V and GND).
• Button on the camera arm where a fingertip finds it. Slide switch under the battery arm.
• Vent slots near the ESP32-CAM; it gets warm.
• PETG preferred (flexes instead of snapping); PLA is fine for the demo. Fixed arms, no hinges.
• Versions: v1 Tue (fit), v2 Wed (pockets and mount), v3 Thu (only if needed).

Risks and backup plans