shoaibahmed912's picture
initial commit
89e6ffa verified
|
Raw History Blame Contribute Delete
3.45 kB
---
title: Transcript Extraction Backend
emoji: 📄
colorFrom: blue
colorTo: green
sdk: docker
app_port: 7860
---
# Transcript Extraction Backend
FastAPI backend that accepts transcript uploads (`jpg`, `jpeg`, `jpp`, `png`, or `pdf`) and returns a structured JSON object with personal details, semesters, subjects, marks, credit hours, GPA, and CGPA where available.
The extraction path is deployable and free-tier friendly:
- FastAPI receives a transcript file
- Gemini reads the uploaded PDF/image directly
- Gemini returns the normalized JSON response
## Setup
```bash
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
```
Create a local `.env` file:
```env
GEMINI_API_KEY=your_gemini_api_key_here
GEMINI_MODEL=gemini-2.5-flash
FIREBASE_SERVICE_ACCOUNT_BASE64=base64_encoded_firebase_service_account_json
TEST_AUTH_TOKEN=local-test-token
```
## Run
```bash
uvicorn app.main:app --reload
```
Open:
```text
http://127.0.0.1:8000/docs
```
## Extract Transcript
```bash
curl -X POST "http://127.0.0.1:8000/extract" \
-H "Authorization: Bearer FIREBASE_ID_TOKEN_FROM_MOBILE_APP" \
-F "file=@/path/to/transcript.pdf"
```
## Firebase Authentication
The mobile app signs the user in with Firebase Auth, gets the user's ID token, and sends it to this backend:
```text
Authorization: Bearer <firebase_id_token>
```
The backend verifies that token with Firebase Admin SDK before processing `/extract`.
For local Swagger testing only, set `TEST_AUTH_TOKEN` in `.env`, restart the server, and enter the same value in the `test_auth_token` parameter shown on `/extract`. Do not set `TEST_AUTH_TOKEN` in production.
To configure the backend, create a Firebase service account:
1. Open Firebase Console.
2. Go to Project settings.
3. Open Service accounts.
4. Generate a new private key.
5. Convert the JSON file to base64 and put it into `FIREBASE_SERVICE_ACCOUNT_BASE64` in `.env` locally or in your cloud host environment variables.
On macOS:
```bash
base64 -i firebase-service-account.json
```
Do not commit the service account JSON file or paste it into mobile app code.
## Free Cloud LLM
For deployment, use the Gemini API free tier. Create an API key in Google AI Studio, then set it as an environment variable in your cloud host:
```bash
export GEMINI_API_KEY=your_api_key_here
export GEMINI_MODEL=gemini-2.5-flash
uvicorn app.main:app --reload
```
The backend sends the uploaded transcript file directly to Gemini and asks it to normalize the transcript into the response JSON schema. If Gemini is not configured, rate-limited, or returns invalid JSON, the API returns an error.
## Deploy On Hugging Face Spaces
This repo includes a `Dockerfile` for Hugging Face Spaces deployment.
Create a new Space:
1. Go to `https://huggingface.co/spaces`.
2. Click `Create new Space`.
3. Choose a name, for example `transcript-data-backend`.
4. Choose `Docker` as the SDK.
5. Choose `CPU Basic`.
6. Create the Space.
7. Upload or push this repository to the Space repo.
Add these as Space secrets:
```env
GEMINI_API_KEY=your_gemini_key
GEMINI_MODEL=gemini-2.5-flash
FIREBASE_SERVICE_ACCOUNT_BASE64=base64_encoded_firebase_service_account_json
```
Do not set `TEST_AUTH_TOKEN` in production.
After the Space builds, open:
```text
https://huggingface.co/spaces/YOUR_USERNAME/YOUR_SPACE_NAME
```
The API docs should be available at:
```text
https://YOUR_USERNAME-YOUR_SPACE_NAME.hf.space/docs
```