slip_vision.py

"""
Gemini vision/OCR helper for brother's 2-odds bet slips.

Given raw image bytes (Telegram photo) and optional caption text, returns
an extracted slip dict compatible with the /slip handler:

{
  "description": "...",
  "stake": float | None,
  "total_odds": float | None,
  "legs": [
    {
      "fixture_hint": str,
      "league_hint": str | None,
      "market": str | None,
      "line": float | None,
      "odds": float | None,
    },
    ...
  ],
}
"""

from __future__ import annotations

import json
import os
from typing import Any

from dotenv import load_dotenv

_script_dir = os.path.dirname(os.path.abspath(__file__))
load_dotenv(os.path.join(_script_dir, ".env"))

GEMINI_API_KEY = (os.getenv("GEMINI_API_KEY") or "").strip()


def _safe_float(val: Any) -> float | None:
    try:
        if val is None:
            return None
        return float(val)
    except (TypeError, ValueError):
        return None


def parse_slip_from_image(image_bytes: bytes, caption: str | None = None) -> dict:
    """Use Gemini multimodal model to parse a bet slip image into a structured slip dict."""
    if not GEMINI_API_KEY or not image_bytes:
        return {}
    try:
        import google.generativeai as genai

        genai.configure(api_key=GEMINI_API_KEY)
        model = genai.GenerativeModel("gemini-2.5-flash")

        img_part = {
            "mime_type": "image/jpeg",
            "data": image_bytes,
        }
        caption_text = (caption or "").strip()
        prompt = (
            "You are reading a sports bet slip image. Extract only structured data.\n\n"
            "Return STRICT JSON ONLY (no prose, no markdown, no code fences) with this schema:\n"
            "{\n"
            '  \"stake\": float or null,\n'
            '  \"total_odds\": float or null,\n'
            '  \"legs\": [\n'
            "    {\n"
            '      \"fixture_hint\": string,          // e.g. \"Liverpool vs Chelsea\"\n'
            '      \"league_hint\": string or null,  // e.g. \"Premier League\"\n'
            '      \"market\": string or null,       // e.g. \"OVER_2_5\", \"HOME_WIN\"\n'
            '      \"line\": float or null,          // e.g. 2.5 for totals\n'
            '      \"odds\": float or null           // per-leg odds if visible\n'
            "    },\n"
            "    ...\n"
            "  ]\n"
            "}\n\n"
            "If something is not visible, set it to null. Do NOT include comments.\n"
        )
        contents: list[Any] = [img_part, {"text": prompt}]
        if caption_text:
            contents.append({"text": f"Caption: {caption_text}"})

        resp = model.generate_content(contents)
        raw = (resp.text or "").strip() if resp else ""
        if not raw:
            return {}

        # Some models may wrap JSON in fences or prose; try to locate the first JSON object.
        raw_str = raw
        if "```" in raw_str:
            # take content between first pair of fences if present
            parts = raw_str.split("```")
            if len(parts) >= 3:
                raw_str = parts[1]
        raw_str = raw_str.strip()

        data = json.loads(raw_str)
        slip: dict = {}
        slip["stake"] = _safe_float(data.get("stake"))
        slip["total_odds"] = _safe_float(data.get("total_odds"))
        legs_out: list[dict] = []
        for leg in data.get("legs") or []:
            if not isinstance(leg, dict):
                continue
            fixture_hint = (leg.get("fixture_hint") or "").strip()
            league_hint = (leg.get("league_hint") or "") or None
            market = (leg.get("market") or "") or None
            line_val = _safe_float(leg.get("line"))
            odds_val = _safe_float(leg.get("odds"))
            if not fixture_hint and not league_hint and not market:
                continue
            legs_out.append(
                {
                    "fixture_hint": fixture_hint,
                    "league_hint": league_hint,
                    "market": market,
                    "line": line_val,
                    "odds": odds_val,
                }
            )
        slip["legs"] = legs_out
        # Description is for human reading; prefer caption if present, else join legs.
        if caption_text:
            slip["description"] = caption_text
        elif legs_out:
            slip["description"] = "; ".join(l.get("fixture_hint") or "" for l in legs_out if l.get("fixture_hint"))
        else:
            slip["description"] = ""
        return slip
    except Exception:
        return {}