Gemini 2.5 Pro × FastAPI: Building a Production-Ready AI Backend

# Create project directory
mkdir gemini-fastapi-backend && cd gemini-fastapi-backend
 
# Create and activate virtual environment
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate
 
# Install dependencies
pip install fastapi uvicorn google-generativeai tenacity python-dotenv pydantic redis

# .env
GEMINI_API_KEY=your_api_key_here
REDIS_URL=redis://localhost:6379
ENVIRONMENT=production
MAX_RETRIES=3

# app/config.py
from pydantic_settings import BaseSettings
 
class Settings(BaseSettings):
    gemini_api_key: str
    environment: str = "development"
    max_retries: int = 3
    redis_url: str = "redis://localhost:6379"
 
    # Gemini model settings
    default_model: str = "gemini-2.5-pro"
    max_output_tokens: int = 8192
    temperature: float = 0.7
 
    class Config:
        env_file = ".env"
 
settings = Settings()

# app/services/gemini.py
import asyncio
import google.generativeai as genai
from tenacity import (
    retry,
    stop_after_attempt,
    wait_exponential,
    retry_if_exception_type,
)
from google.api_core.exceptions import ResourceExhausted, ServiceUnavailable
from app.config import settings
import logging
 
logger = logging.getLogger(__name__)
 
genai.configure(api_key=settings.gemini_api_key)
 
class GeminiClient:
    def __init__(self, model_name: str = None):
        self.model_name = model_name or settings.default_model
        self.generation_config = genai.types.GenerationConfig(
            max_output_tokens=settings.max_output_tokens,
            temperature=settings.temperature,
        )
        self.model = genai.GenerativeModel(
            model_name=self.model_name,
            generation_config=self.generation_config,
        )
 
    @retry(
        stop=stop_after_attempt(settings.max_retries),
        wait=wait_exponential(multiplier=1, min=4, max=60),
        retry=retry_if_exception_type((ResourceExhausted, ServiceUnavailable)),
        before_sleep=lambda retry_state: logger.warning(
            f"Rate limited. Retrying in {retry_state.next_action.sleep}s "
            f"(attempt {retry_state.attempt_number}/{settings.max_retries})"
        ),
    )
    async def generate(self, prompt: str, system_instruction: str = None) -> str:
        """
        Content generation with exponential backoff retry.
        Automatically retries on ResourceExhausted (429) or ServiceUnavailable (503).
        """
        model = self.model
        if system_instruction:
            model = genai.GenerativeModel(
                model_name=self.model_name,
                generation_config=self.generation_config,
                system_instruction=system_instruction,
            )
 
        # Use asyncio.to_thread to run the blocking SDK call asynchronously
        response = await asyncio.to_thread(
            model.generate_content, prompt
        )
        return response.text
 
    async def stream_generate(self, prompt: str, system_instruction: str = None):
        """
        Streaming generation. Yields each chunk as an async generator.
        """
        model = self.model
        if system_instruction:
            model = genai.GenerativeModel(
                model_name=self.model_name,
                generation_config=self.generation_config,
                system_instruction=system_instruction,
            )
 
        response = await asyncio.to_thread(
            model.generate_content, prompt, stream=True
        )
 
        for chunk in response:
            if chunk.text:
                yield chunk.text
 
gemini_client = GeminiClient()

# app/routers/stream.py
from fastapi import APIRouter, HTTPException
from fastapi.responses import StreamingResponse
from pydantic import BaseModel
from app.services.gemini import gemini_client
import json
import asyncio
 
router = APIRouter(prefix="/api/v1", tags=["streaming"])
 
class StreamRequest(BaseModel):
    message: str
    system_instruction: str | None = None
    session_id: str | None = None
 
async def event_generator(request: StreamRequest):
    """
    Generates streaming responses in Server-Sent Events (SSE) format.
    Each chunk is sent as: {"type": "chunk", "content": "..."}
    Errors are sent as: {"type": "error", "message": "..."}
    Completion is signaled by: {"type": "done"}
    """
    try:
        async for chunk in gemini_client.stream_generate(
            prompt=request.message,
            system_instruction=request.system_instruction,
        ):
            data = json.dumps({"type": "chunk", "content": chunk})
            yield f"data: {data}\n\n"
            # Yield control to prevent buffer overflow
            await asyncio.sleep(0)
 
        yield f"data: {json.dumps({'type': 'done'})}\n\n"
 
    except Exception as e:
        error_data = json.dumps({"type": "error", "message": str(e)})
        yield f"data: {error_data}\n\n"
 
@router.post("/stream")
async def stream_chat(request: StreamRequest):
    """
    Streaming chat endpoint using Server-Sent Events.
 
    Example client-side usage (JavaScript):
    const eventSource = new EventSource('/api/v1/stream');
    eventSource.onmessage = (e) => {
        const data = JSON.parse(e.data);
        if (data.type === 'chunk') process.stdout.write(data.content);
    };
    """
    if not request.message.strip():
        raise HTTPException(status_code=400, detail="Message cannot be empty")
 
    return StreamingResponse(
        event_generator(request),
        media_type="text/event-stream",
        headers={
            "Cache-Control": "no-cache",
            "X-Accel-Buffering": "no",  # Disable Nginx buffering
        },
    )

# app/services/tools.py
import google.generativeai as genai
import asyncio
 
TOOLS = [
    genai.protos.Tool(
        function_declarations=[
            genai.protos.FunctionDeclaration(
                name="get_current_weather",
                description="Get current weather information for a specified city",
                parameters=genai.protos.Schema(
                    type=genai.protos.Type.OBJECT,
                    properties={
                        "city": genai.protos.Schema(
                            type=genai.protos.Type.STRING,
                            description="City name (e.g., Tokyo, New York)",
                        ),
                        "unit": genai.protos.Schema(
                            type=genai.protos.Type.STRING,
                            enum=["celsius", "fahrenheit"],
                            description="Temperature unit",
                        ),
                    },
                    required=["city"],
                ),
            ),
            genai.protos.FunctionDeclaration(
                name="search_articles",
                description="Search for articles on Gemini Lab",
                parameters=genai.protos.Schema(
                    type=genai.protos.Type.OBJECT,
                    properties={
                        "query": genai.protos.Schema(
                            type=genai.protos.Type.STRING,
                            description="Search query",
                        ),
                        "limit": genai.protos.Schema(
                            type=genai.protos.Type.INTEGER,
                            description="Number of articles to return (default: 5)",
                        ),
                    },
                    required=["query"],
                ),
            ),
        ]
    )
]
 
async def execute_tool(function_name: str, args: dict) -> dict:
    """Execute the tool called by Function Calling"""
    if function_name == "get_current_weather":
        city = args.get("city", "Tokyo")
        unit = args.get("unit", "celsius")
        # In production, call a real weather API like Open-Meteo
        return {
            "city": city,
            "temperature": 22,
            "unit": unit,
            "condition": "Sunny",
            "humidity": 55,
        }
 
    elif function_name == "search_articles":
        query = args.get("query", "")
        limit = args.get("limit", 5)
        return {
            "results": [
                {
                    "title": f"Complete Guide to Gemini {query}",
                    "url": f"(/articles/gemini-api/{query.lower().replace(' ', '-')}",
                }
            ],
            "total": 1,
        }
 
    return {"error": f"Unknown function: {function_name}"}

# app/routers/chat.py
from fastapi import APIRouter, HTTPException
from pydantic import BaseModel
import google.generativeai as genai
import asyncio
from app.config import settings
from app.services.tools import TOOLS, execute_tool
import logging
 
router = APIRouter(prefix="/api/v1", tags=["chat"])
logger = logging.getLogger(__name__)
 
class ChatRequest(BaseModel):
    message: str
    history: list[dict] = []
    use_tools: bool = True
 
class ChatResponse(BaseModel):
    response: str
    tool_calls: list[dict] = []
    model: str
    tokens_used: int | None = None
 
@router.post("/chat", response_model=ChatResponse)
async def chat(request: ChatRequest):
    """
    Chat endpoint with Function Calling support.
    Executes multi-turn reasoning while calling external tools as needed.
 
    Expected response:
    {
      "response": "The weather in Tokyo is currently 22°C and sunny.",
      "tool_calls": [{"function": "get_current_weather", "args": {"city": "Tokyo"}}],
      "model": "gemini-2.5-pro",
      "tokens_used": 342
    }
    """
    genai.configure(api_key=settings.gemini_api_key)
 
    model = genai.GenerativeModel(
        model_name=settings.default_model,
        tools=TOOLS if request.use_tools else None,
    )
 
    chat_history = [
        {"role": msg["role"], "parts": [msg["content"]]}
        for msg in request.history
    ]
 
    chat_session = model.start_chat(history=chat_history)
    tool_calls_log = []
 
    response = await asyncio.to_thread(chat_session.send_message, request.message)
 
    # Function Calling loop
    while (
        response.candidates and
        response.candidates[0].content.parts and
        hasattr(response.candidates[0].content.parts[0], 'function_call') and
        response.candidates[0].content.parts[0].function_call.name
    ):
        fc = response.candidates[0].content.parts[0].function_call
        func_name = fc.name
        func_args = dict(fc.args)
 
        logger.info(f"Tool call: {func_name}({func_args})")
        tool_calls_log.append({"function": func_name, "args": func_args})
 
        tool_result = await execute_tool(func_name, func_args)
 
        response = await asyncio.to_thread(
            chat_session.send_message,
            genai.protos.Part(
                function_response=genai.protos.FunctionResponse(
                    name=func_name,
                    response={"result": tool_result},
                )
            )
        )
 
    tokens_used = None
    if hasattr(response, 'usage_metadata'):
        tokens_used = response.usage_metadata.total_token_count
 
    return ChatResponse(
        response=response.text,
        tool_calls=tool_calls_log,
        model=settings.default_model,
        tokens_used=tokens_used,
    )

# app/main.py
from fastapi import FastAPI
from fastapi.middleware.cors import CORSMiddleware
from contextlib import asynccontextmanager
import logging
from app.routers import chat, stream
from app.config import settings
 
logging.basicConfig(
    level=logging.INFO if settings.environment == "production" else logging.DEBUG,
    format="%(asctime)s - %(name)s - %(levelname)s - %(message)s",
)
 
@asynccontextmanager
async def lifespan(app: FastAPI):
    logging.info(f"🚀 Starting Gemini FastAPI Backend (env: {settings.environment})")
    yield
    logging.info("🛑 Shutting down Gemini FastAPI Backend")
 
app = FastAPI(
    title="Gemini FastAPI Backend",
    description="Production-ready AI backend powered by Gemini 2.5 Pro × FastAPI",
    version="1.0.0",
    lifespan=lifespan,
)
 
app.add_middleware(
    CORSMiddleware,
    allow_origins=["https://yourdomain.com"] if settings.environment == "production" else ["*"],
    allow_credentials=True,
    allow_methods=["*"],
    allow_headers=["*"],
)
 
app.include_router(chat.router)
app.include_router(stream.router)
 
@app.get("/health")
async def health_check():
    """Health check endpoint for load balancers"""
    return {"status": "ok", "model": settings.default_model}
 
# Expected output (curl http://localhost:8000/health):
# {"status": "ok", "model": "gemini-2.5-pro"}

# app/middleware/rate_limit.py
import asyncio
import time
 
class AdaptiveRateLimiter:
    """
    Adaptive rate limiter that automatically widens request intervals
    when 429 errors increase, and tightens them as successes accumulate.
    """
    def __init__(self, initial_rps: float = 10.0):
        self.min_interval = 1.0 / initial_rps
        self.current_interval = self.min_interval
        self.last_request_time = 0.0
        self.lock = asyncio.Lock()
        self._consecutive_errors = 0
        self._consecutive_successes = 0
 
    async def acquire(self):
        async with self.lock:
            now = time.monotonic()
            elapsed = now - self.last_request_time
            if elapsed < self.current_interval:
                await asyncio.sleep(self.current_interval - elapsed)
            self.last_request_time = time.monotonic()
 
    def record_success(self):
        self._consecutive_successes += 1
        self._consecutive_errors = 0
        if self._consecutive_successes >= 10:
            self.current_interval = max(self.min_interval, self.current_interval * 0.9)
            self._consecutive_successes = 0
 
    def record_rate_limit(self):
        self._consecutive_errors += 1
        self._consecutive_successes = 0
        self.current_interval = min(60.0, self.current_interval * 2.0)
 
rate_limiter = AdaptiveRateLimiter(initial_rps=5.0)

# app/services/cache.py
import hashlib
import json
from datetime import timedelta
import redis.asyncio as redis
from app.config import settings
 
class ResponseCache:
    def __init__(self):
        self.client = redis.from_url(settings.redis_url)
        self.ttl = timedelta(hours=24)
 
    def _make_key(self, prompt: str, model: str) -> str:
        content = f"{model}:{prompt}"
        return f"gemini:cache:{hashlib.sha256(content.encode()).hexdigest()}"
 
    async def get(self, prompt: str, model: str) -> str | None:
        key = self._make_key(prompt, model)
        cached = await self.client.get(key)
        return json.loads(cached) if cached else None
 
    async def set(self, prompt: str, model: str, response: str):
        key = self._make_key(prompt, model)
        await self.client.setex(
            key,
            int(self.ttl.total_seconds()),
            json.dumps(response)
        )
 
cache = ResponseCache()

async def generate_with_timeout(prompt: str, timeout: float = 120.0) -> str:
    try:
        return await asyncio.wait_for(
            asyncio.to_thread(model.generate_content, prompt),
            timeout=timeout
        )
    except asyncio.TimeoutError:
        raise HTTPException(status_code=504, detail="API timeout (120s exceeded)")

FROM python:3.11-slim
 
WORKDIR /app
 
# Security: run as non-root user
RUN addgroup --system appgroup && adduser --system --ingroup appgroup appuser
 
COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt
 
COPY app/ ./app/
 
USER appuser
 
HEALTHCHECK --interval=30s --timeout=10s --start-period=5s --retries=3 \
    CMD curl -f http://localhost:8000/health || exit 1
 
EXPOSE 8000
 
# Production: recommended workers = (CPU count × 2) + 1
CMD ["uvicorn", "app.main:app", "--host", "0.0.0.0", "--port", "8000", "--workers", "4"]

version: "3.9"
 
services:
  api:
    build: .
    ports:
      - "8000:8000"
    environment:
      - GEMINI_API_KEY=${GEMINI_API_KEY}
      - REDIS_URL=redis://redis:6379
      - ENVIRONMENT=production
    depends_on:
      redis:
        condition: service_healthy
    restart: unless-stopped
 
  redis:
    image: redis:7-alpine
    healthcheck:
      test: ["CMD", "redis-cli", "ping"]
      interval: 5s
      timeout: 3s
      retries: 5
    restart: unless-stopped
 
# Launch with:
# GEMINI_API_KEY=your_key docker-compose up -d