fastapi-translation

English → Turkish translator API built with FastAPI and Hugging Face models (ckartal/english-to-turkish-finetuned-model).

App identity (Docker & AWS)

All resources use the name fastapi-translation so you can trace what runs where:

Environment	Image	Container / service
Local Docker	fastapi-translation:local	fastapi-translation-app
AWS ECR	{account}.dkr.ecr.../fastapi-translation:latest	ECS container fastapi-translation
AWS ECS	—	cluster fastapi-translation-cluster, service fastapi-translation-service

Endpoints

Path	Description
/	Web UI
/health	Health check ({"status":"ok","app":"fastapi-translation"})
/apidocs	Swagger UI
POST /translate/	Translate JSON {"text":"..."}
POST /translate-and-speak/	Translate + WAV audio

Quick start (local Docker)

chmod +x scripts/*.sh scripts/lib/common.sh
./scripts/local.sh
./scripts/test-api.sh http://localhost:3000

Stop: ./scripts/stop-local.sh

First boot downloads models into .cache/huggingface (several minutes).

AWS ECS deploy (Terraform)

Docs: GUIDE.md · TERRAFORM.md · PROD.md

cp .env.example .env
export AWS_PROFILE=terraform-user
aws configure --profile terraform-user

./scripts/plan.sh dev
./scripts/deploy.sh dev    # plan → yes → apply → push image → ECS
./scripts/status.sh

Goal	Command
Local	./scripts/local.sh
Preview infra	./scripts/plan.sh dev
Full deploy	./scripts/deploy.sh dev
App code only	./scripts/redeploy-app.sh dev
Prod	./scripts/deploy.sh prod
Destroy	./scripts/destroy.sh dev

Detected config: project fastapi-translation, port 3000, health /health, Python/FastAPI, ARM64 Fargate.

Local Python (without Docker)

pip install -r requirements-docker.txt
pip install torch --index-url https://download.pytorch.org/whl/cpu
uvicorn main:app --host 0.0.0.0 --port 3000

Vercel

Lightweight index.py (HTTP translation API) is deployed via vercel.json — separate from the Docker/ECS ML stack in main.py.