Project Description
I want a working prototype of a truly pocket-sized gadget that can run completely offline. The flow is straightforward: a standard-resolution camera captures a printed or handwritten question, an on-board OCR module extracts the text, a lightweight language model generates a full textual answer, and the reply appears on a small built-in screen. No cloud calls, no external power brick—everything must live inside the device itself.
I am open on the exact OCR library; Tesseract OCR is the obvious first choice, but if you have a lighter or faster alternative that will compile on something like a Raspberry Pi Zero 2 W, an ESP32-S3, or any comparable single-board solution, I’m happy to discuss it. For the AI model, please keep it compact (e.g., TensorFlow Lite, ONNX, or similar) so that inference happens locally without noticeable lag.
The prototype should fit comfortably in a trouser pocket, run off a small rechargeable battery, and boot to answer in under ten seconds. Latency from camera capture to answer on screen should not exceed five seconds for a single printed sentence.
Deliverables
• Fully assembled working prototype (camera, processing board, battery, display, enclosure)
• All source code with build instructions (C/C++, Python, or MicroPython)
• PCB layout files and 3-D enclosure files (STL or STEP)
• Short demonstration video showing the device answering at least three sample questions offline
• Brief documentation: component list, wiring diagram, model size, power draw, and instructions for flashing or updating the firmware
When handing over, make sure every step—from image capture with OpenCV to text extraction in the OCR layer to answer generation in the embedded AI model—can be replicated on a fresh board using only the files you provide.