Skip to content
Rakha Yuda Wicaksono
← All projects

2026 · Full-Stack

Computer Vision for FTTH Sales Data Validation

Computer-vision and FastAPI system that validates FTTH sales data, verifies sales visits, and confirms customers by OTP.

Private build

Problem

The previous version of the app was hard to use and its flow was confusing. It had no way to request an extra FAT pole, no AI to verify that a sales visit really happened, and no validation of customer identity, so the reported data could not be trusted.

Role & constraints

Solo full-stack work from March to August 2026, as my final-year thesis. The code is private. I owned the backend, the Flutter app, the PostgreSQL schema, and the computer-vision model at the same time.

Approach

I added OTP-based customer validation and a computer-vision module built on YOLO26 and PaddleOCR that detects and reads visit evidence, so a visit is confirmed by the system instead of by the salesperson's word. I simplified the business flow, added a request path for extra FAT poles, and rebuilt the interface in Flutter. FastAPI serves the API, PostgreSQL stores the data, and Redis handles caching for the reads that repeat most.

Result

Visits are verified automatically and customers are confirmed with OTP, which removes the manual spot-checks. The hardest part was carrying backend, frontend, and the AI model in one project.