# app.py (Hugging Face Space)

from fastapi import FastAPI, HTTPException
from pydantic import BaseModel

# ---- import your real logic ----
from model.semantic.skills import extract_skills
from model.semantic.embeddings import embed_texts
from model.semantic.similarity import semantic_similarity
from model.semantic.scoring import compute_scores
from model.semantic.jd import prepare_jd_chunks

from model.llm.prompt import build_prompt
from model.llm.client import run_llm
from model.llm.parser import parse_llm_output

from model.bias.detector import run_bias_rules
from model.bias.prompt import build_bias_prompt
from model.bias.parser import parse_bias_output

app = FastAPI()

# -------- Request schema --------
class ResumeRequest(BaseModel):
    resume_text: str
    resume_sections: dict
    job_description: str

# -------- Health check --------
@app.get("/health")
def health():
    return {"status": "ok"}

# -------- Inference endpoint --------
@app.post("/analyze")
def analyze_resume(req: ResumeRequest):
    structured_resume = req.resume_sections

    resume_skills = extract_skills(structured_resume.get("skills", []))
    jd_skills = extract_skills([req.job_description])

    resume_chunks = (
        structured_resume.get("experience", [])
        + structured_resume.get("projects", [])
    )

    if not resume_chunks:
        raise HTTPException(status_code=400, detail="No experience or projects")

    resume_vecs = embed_texts(resume_chunks)
    jd_vecs = embed_texts(prepare_jd_chunks(req.job_description))

    semantic_score = semantic_similarity(resume_vecs, jd_vecs)
    scores = compute_scores(resume_skills, jd_skills, semantic_score)

    try:
        prompt = build_prompt(
            structured_resume=structured_resume,
            skills_found=list(resume_skills),
            scores=scores,
            job_description=req.job_description
        )
        explanation = parse_llm_output(run_llm(prompt))
    except Exception as e:
        explanation = {"error": str(e)}

    bias_findings = run_bias_rules(
        resume_text=req.resume_text,
        jd_text=req.job_description,
        scores=scores
    )

    if bias_findings:
        bias_report = parse_bias_output(
            run_llm(build_bias_prompt(bias_findings))
        )
    else:
        bias_report = {
            "detected_biases": [],
            "risk_level": "low",
            "explanations": [],
            "mitigation_suggestions": []
        }

    return {
        "scores": scores,
        "semantic_score": semantic_score,
        "explanation": explanation,
        "bias_report": bias_report
    }