Client:
Kaizen
Tech stack:
Python, FastAPI, MongoDB, JavaScript, HTML5, CSS
Challenge:
Achieving accurate OCR results was difficult, as initial prompts caused confusion between multiple roles. Additionally, handling location-based data added further complexity.
Solution:
Researched various Medium blogs and implemented a three-prompt strategy, assigning a distinct role to each prompt to prevent confusion and improve data accuracy.
Outcome:
- Accuracy improved to beyond 90–95%, significantly enhancing data extraction precision. The client successfully proceeded with the implementation of our OCR software.
- Most Challenging Project: AI-OCR – due to the complexity of handling location-based data.
- Innovation Spotlight: Developed a multi-prompt strategy by assigning distinct roles to each prompt, reducing confusion and significantly improving OCR accuracy beyond conventional methods.