"""FastAPI service layer for the AI Video Intelligence Agent. Required endpoints GET /health GET /metrics POST /video/analyze GET /video/{video_id}/timeline GET /video/{video_id}/bookmarks POST /video/search Optional endpoints POST /video/ask GET /video/{video_id}/summary GET /video/{video_id}/transcript GET /video/{video_id}/visual-events In ``DEMO_MODE`` / ``USE_PRECOMPUTED`` (the defaults) the service reads the committed sample artifacts and never loads a heavy model. Live analysis only runs when precomputed mode is disabled *and* the local ML deps are installed. Run locally: uvicorn api.server:app --reload --port 8000 """ from __future__ import annotations import tempfile from pathlib import Path from typing import List, Optional from fastapi import FastAPI, File, Form, HTTPException, UploadFile from pydantic import BaseModel from src import storage from src.config import CONFIG from src.metrics import latency_summary from src.search import VideoSearch from src.summarize import answer_question, extractive_summary from src.visual_analysis import object_frequency app = FastAPI( title="AI Video Intelligence Agent", description=( "Multimodal analysis of short learning videos: visual events, " "Thai-English ASR, searchable timeline, automatic bookmarks." ), version="0.1.0", ) # --------------------------------------------------------------------------- # # Request models # --------------------------------------------------------------------------- # class SearchRequest(BaseModel): video_id: str query: str top_k: int = 10 class AskRequest(BaseModel): video_id: str question: str top_k: int = 3 def _require_timeline(video_id: str) -> dict: timeline = storage.load_timeline(video_id) if timeline is None: raise HTTPException(status_code=404, detail=f"Unknown video_id '{video_id}'.") return timeline # --------------------------------------------------------------------------- # # Core endpoints # --------------------------------------------------------------------------- # @app.get("/health") def health() -> dict: return { "status": "ok", "demo_mode": CONFIG.demo_mode, "use_precomputed": CONFIG.use_precomputed, "available_videos": storage.list_sample_video_ids(), } @app.get("/metrics") def metrics() -> dict: """Aggregate processing + search metrics across all available videos.""" videos = {} all_latencies: List[float] = [] for vid in storage.list_sample_video_ids(): m = storage.load_metrics(vid) if not m: continue videos[vid] = m sl = (m.get("search_latency") or {}) # Reconstruct latency points isn't stored; surface the summary instead. if sl.get("p50_ms") is not None: all_latencies.append(sl["p50_ms"]) return { "n_videos": len(videos), "videos": videos, "search_latency_p50_across_videos_ms": ( latency_summary(all_latencies)["p50_ms"] if all_latencies else None ), "config": CONFIG.public_dict(), } @app.post("/video/analyze") async def analyze( file: Optional[UploadFile] = File(default=None), url: Optional[str] = Form(default=None), video_id: Optional[str] = Form(default=None), ) -> dict: """Analyze an uploaded video / a URL, or return a precomputed analysis by id. * No file/url + known ``video_id`` -> returns the precomputed artifacts. * File or URL, precomputed mode -> returns a note (demo stays stable). * File or URL, live mode -> runs the local pipeline (needs ML deps). ``url`` may point to YouTube, Vimeo, a direct link, etc. (handled by yt-dlp). """ if file is None and not url: if not video_id: raise HTTPException( status_code=400, detail="Provide a video file, a 'url', or a known 'video_id'.", ) timeline = _require_timeline(video_id) return { "video_id": video_id, "mode": "precomputed", "timeline": timeline, "bookmarks": storage.load_bookmarks(video_id), "metrics": storage.load_metrics(video_id), } if CONFIG.use_precomputed or CONFIG.demo_mode: return { "mode": "demo", "message": ( "Demo/precomputed mode is on — live analysis (file upload and " "URL ingestion) is disabled for stability. Set " "USE_PRECOMPUTED=false and DEMO_MODE=false to enable local " "analysis, or query a precomputed sample." ), "available_videos": storage.list_sample_video_ids(), } # ---- live analysis path -------------------------------------------- # try: from src.pipeline import analyze_video # lazy: pulls heavy deps except Exception as exc: # pragma: no cover raise HTTPException( status_code=503, detail=f"Local analysis dependencies unavailable: {exc}", ) tmp_path: Optional[Path] = None try: if file is not None: suffix = Path(file.filename or "upload.mp4").suffix or ".mp4" with tempfile.NamedTemporaryFile(delete=False, suffix=suffix) as tmp: tmp.write(await file.read()) tmp_path = Path(tmp.name) source: object = tmp_path else: source = url # URL string -> resolved/downloaded by the pipeline artifacts = analyze_video(source, video_id=video_id, config=CONFIG) except (ValueError, FileNotFoundError) as exc: raise HTTPException(status_code=400, detail=str(exc)) finally: if tmp_path is not None: tmp_path.unlink(missing_ok=True) return { "video_id": artifacts["metrics"]["video_id"], "mode": "live", "timeline": artifacts["timeline"], "bookmarks": artifacts["bookmarks"], "metrics": artifacts["metrics"], } @app.get("/video/{video_id}/timeline") def get_timeline(video_id: str) -> dict: return _require_timeline(video_id) @app.get("/video/{video_id}/bookmarks") def get_bookmarks(video_id: str) -> List[dict]: bookmarks = storage.load_bookmarks(video_id) if bookmarks is None: raise HTTPException(status_code=404, detail=f"Unknown video_id '{video_id}'.") return bookmarks @app.post("/video/search") def search(req: SearchRequest) -> dict: timeline = _require_timeline(req.video_id) return VideoSearch(timeline, CONFIG).search(req.query, top_k=req.top_k) # --------------------------------------------------------------------------- # # Optional endpoints # --------------------------------------------------------------------------- # @app.post("/video/ask") def ask(req: AskRequest) -> dict: timeline = _require_timeline(req.video_id) return answer_question(req.question, timeline, CONFIG, top_k=req.top_k) @app.get("/video/{video_id}/summary") def get_summary(video_id: str) -> dict: timeline = _require_timeline(video_id) return extractive_summary(timeline, CONFIG) @app.get("/video/{video_id}/transcript") def get_transcript(video_id: str) -> List[dict]: transcript = storage.load_transcript(video_id) if transcript is None: raise HTTPException(status_code=404, detail=f"Unknown video_id '{video_id}'.") return transcript @app.get("/video/{video_id}/visual-events") def get_visual_events(video_id: str) -> dict: visual = storage.load_visual_events(video_id) if visual is None: raise HTTPException(status_code=404, detail=f"Unknown video_id '{video_id}'.") return { "video_id": video_id, "n_frames": len(visual), "object_frequency": object_frequency(visual), "frames": visual, }