Our operation photographs every retreaded tire sidewall in a controlled light box with a fixed-position DSLR, producing crisp JPEG frames. I need a compact pipeline that can take each image, read every character on the sidewall, and hand back a clean text file with the fields that matter to us: brand, model, full DOT code, and any extra markings we specify in a short reference sheet I will share on kickoff. What matters most is accuracy. I have to reach roughly 99 % character-level recognition across a validation set; without that, the downstream quality checks fail, so your code must include whatever pre-processing, language models, or post-correction logic is required to hit that mark. The job is two-fold—first the OCR itself, then the logic that parses the raw text into the structured fields; both pieces belong in the final deliverable. Please keep the workflow straightforward: drop a batch of JPEGs into a folder, run one command, and receive a matching set of .txt files. Python with Tesseract, EasyOCR, OpenCV, or a tuned deep-learning model is fine so long as the dependencies are documented and install in one shot on Ubuntu. Deliverables • Source code and requirements file • A runnable script or notebook that accepts a folder of JPEGs and outputs per-image .txt files with the required fields in plain text • Short README covering setup, usage, and the steps you applied to reach the 99 % target • A small demo run on five sample images showing the extracted fields Acceptance criteria: the supplied validation images must parse with at least 99 % accuracy for all characters and every mandatory field must be present in the output. I would like everything wrapped up within 14 days, including one round of feedback.