This project focuses on building a hybrid deep learning model to solve captchas used in the SRMAP Student Portal and Parent Portal.
The core of this repository is a CRNN-based captcha recognition model (CNN + LSTM + CTC) trained to work reliably across both portals using a single model. Alongside the model, a production-ready FastAPI service is provided to deploy the solver in real-world systems.
This solution is actively used in production by the srmapi.in website.
- Primary focus: Captcha recognition model
- Secondary component: Production-ready inference API
- One hybrid model supports student + parent portals
- Designed for stability, scalability, and real usage
- CRNN architecture (CNN + BiLSTM + CTC)
- Handles variable-length captcha text
- Robust to noise and distortion
- Single model for multiple captcha formats
A FastAPI-based inference service is included to deploy the model safely.
Key characteristics:
- Queue-based inference
- Single-worker execution (GPU/CPU safe)
- Plain-text responses
- Used in live production traffic
POST /captcha
- Content-Type:
multipart/form-data - Field name:
file - Type: Captcha image (PNG / JPG / JPEG)
curl -X POST http://localhost:6000/captcha \
-F "file=@captcha.png"- Content-Type:
text/plain - Body: Decoded captcha text
A7F9Q
Built by:
Azam
Brahmendra
This project is intended for educational and research purposes related to SRMAP systems.
Use responsibly and in accordance with applicable rules and policies.