api.py

from fastapi import FastAPI, HTTPException
from fastapi.middleware.cors import CORSMiddleware
from fastapi.openapi.docs import get_swagger_ui_html
from pydantic import BaseModel, model_validator
from typing import Optional

from environment import CustomerSupportEnv, STEP_ORDER
from graders.base_grader import BaseGrader, HardTaskGrader
from tasks.easy_task import EASY_TASK
from tasks.medium_task import MEDIUM_TASK
from tasks.hard_task import HARD_TASK

app = FastAPI(title="Customer Support AI Environment")

app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_methods=["*"],
    allow_headers=["*"],
)

_env: Optional[CustomerSupportEnv] = None

TASK_MAP = {
    "easy":   EASY_TASK,
    "medium": MEDIUM_TASK,
    "hard":   HARD_TASK,
}


class ResetRequest(BaseModel):
    task: str = "easy"


class StepRequest(BaseModel):
    action: Optional[str] = None    # accepts {"action": "..."}
    response: Optional[str] = None  # accepts {"response": "..."}

    @model_validator(mode="after")
    def resolve_action(self):
        self.action = self.action or self.response
        if not self.action:
            raise ValueError("Provide either 'action' or 'response' field with the agent reply.")
        return self


def _build_observation():
    ep = _env.episode
    task = _env.task
    step_index = _env._step_index
    current_step_name = (
        STEP_ORDER[step_index].value if step_index < len(STEP_ORDER) else "done"
    )
    return {
        "task_id": task.task_id,
        "difficulty": task.difficulty.value,
        "customer_emotion": task.customer_emotion,
        "customer_message": task.customer_message,
        "scenario_context": task.scenario_context,
        "current_step": current_step_name,
        "step_number": step_index + 1,
        "total_steps": len(STEP_ORDER),
        "episode_status": ep.status.value,
        "total_reward": round(ep.total_reward, 3),
        "wrong_step_count": ep.wrong_step_count,
        "steps_completed": [
            {
                "step": s.step.value,
                "correct": s.correct,
                "detected_action": s.detected_action,
                "reward": round(s.reward, 3),
                "penalty": round(s.penalty, 3),
                "penalty_reasons": s.penalty_reasons,
            }
            for s in ep.steps
        ],
    }


@app.get("/")
def health_check():
    return {"status": "ok", "message": "Customer Support AI Environment"}


@app.post("/reset")
def reset_env(body: ResetRequest = None):
    global _env
    task_key = (body.task if body else "easy").lower()
    if task_key not in TASK_MAP:
        raise HTTPException(
            status_code=400,
            detail=f"Unknown task '{task_key}'. Choose from: {list(TASK_MAP.keys())}"
        )
    task = TASK_MAP[task_key]
    grader = HardTaskGrader() if task_key == "hard" else BaseGrader()
    _env = CustomerSupportEnv(task=task, grader=grader)
    _env.reset()
    return {"observation": _build_observation()}


@app.post("/step")
def step_env(body: StepRequest):
    global _env
    if _env is None:
        raise HTTPException(status_code=400, detail="Environment not initialized. Call /reset first.")
    if _env.episode.status.value != "running":
        raise HTTPException(status_code=400, detail=f"Episode already ended: {_env.episode.status.value}")
    result, done = _env.step(body.action)
    return {
        "observation": _build_observation(),
        "reward": round(result.reward, 3),
        "done": done,
        "info": {
            "step": result.step.value,
            "correct": result.correct,
            "detected_action": result.detected_action,
            "expected_action": result.expected_action,
            "penalty": round(result.penalty, 3),
            "penalty_reasons": result.penalty_reasons,
            "fail_triggered": result.fail_triggered,
            "fail_reason": result.fail_reason,
        },
    }


@app.get("/state")
def get_state():
    if _env is None:
        raise HTTPException(status_code=400, detail="Environment not initialized. Call /reset first.")
    return _env.summary()


@app.get("/observation_space")
def observation_space():
    return {
        "type": "Dict",
        "fields": {
            "task_id": "str",
            "difficulty": "str (easy|medium|hard)",
            "customer_emotion": "str",
            "customer_message": "str",
            "scenario_context": "str",
            "current_step": "str (empathy|collect_info|investigate|resolution|done)",
            "step_number": "int",
            "total_steps": "int",
            "episode_status": "str (running|success|fail)",
            "total_reward": "float",
            "wrong_step_count": "int",
            "steps_completed": "List[Dict]",
        }
    }


@app.get("/action_space")
def action_space():
    return {
        "type": "Text",
        "description": "Agent free-text reply to customer",
        "min_words": 6,
    }


@app.get("/docs", include_in_schema=False)
def custom_docs():
    return get_swagger_ui_html(openapi_url="/openapi.json", title="API Docs")


def main():
    """Entry point for 'server' script defined in pyproject.toml."""
    import uvicorn
    uvicorn.run("api:app", host="0.0.0.0", port=7860)


if __name__ == "__main__":
    main()