from fastapi import FastAPI, UploadFile, File, HTTPException, Query from fastapi.responses import FileResponse from fastapi.staticfiles import StaticFiles from fastapi.middleware.cors import CORSMiddleware from pydantic import BaseModel import tempfile import os import logging import json from live_mode_example import clean_live_text, client # your existing module logger = logging.getLogger("vrs-app") app = FastAPI() # Allow browser JS to call the API app.add_middleware( CORSMiddleware, allow_origins=["*"], allow_methods=["*"], allow_headers=["*"], ) # Serve static files (front-end) app.mount("/static", StaticFiles(directory="static"), name="static") @app.get("/") async def root(): """Serve the main HTML page.""" return FileResponse("static/index.html") @app.post("/api/transcribe") async def transcribe( mode: str = Query("smart", regex="^(direct|smart)$"), specialty: str = Query("radiology"), # Optional, defaults to radiology file: UploadFile = File(...), ): """ mode = "direct" -> return raw transcript only mode = "smart" -> try to return cleaned transcript using clean_live_text() (falls back to raw if cleaner fails) """ # 1) Save uploaded file to temp with tempfile.NamedTemporaryFile(delete=False, suffix=".webm") as tmp: data = await file.read() tmp.write(data) tmp_path = tmp.name try: # 2) Transcription with OpenAI (common for both modes) try: with open(tmp_path, "rb") as f: transcription = client.audio.transcriptions.create( model="gpt-4o-mini-transcribe", file=f, response_format="text", # plain string language="en", # force English ) except Exception as e: logger.exception("OpenAI transcription error: %s", e) raise HTTPException( status_code=500, detail=f"Transcription error: {e}" ) raw_text = ( transcription if isinstance(transcription, str) else getattr(transcription, "text", "") ) if not raw_text.strip(): raise HTTPException( status_code=400, detail="No text recognized from audio." ) cleaned = None clean_error = None # 3) Optional cleaning for SMART mode if mode == "smart": try: cleaned = clean_live_text( raw_transcript=raw_text, specialty=specialty, language="en", ) except Exception as e: clean_error = str(e) cleaned = None logger.exception("clean_live_text failed: %s", e) # do NOT raise; we still return raw text # 4) Respond return { "mode": mode, "raw": raw_text, "cleaned": cleaned, "clean_error": clean_error, } finally: try: os.remove(tmp_path) except OSError: pass # ============================================ # NEW ENDPOINTS: Review, Modify, Generate Report # ============================================ # System prompt for AI tasks AI_SYSTEM_PROMPT = """ You are an AI assistant specialized in healthcare documentation for a Voice Reporting System (VRS). Your role depends on the task: 1. REVIEW (task="review"): - Clean and polish the provided medical text - Fix grammar, punctuation, and capitalization - Ensure proper medical terminology and formatting - Preserve ALL medical meaning, findings, diagnoses, and numbers - Do NOT add or remove medical information - Maintain the original structure and flow 2. MODIFY (task="modify"): - Apply the user's natural language instruction to modify the existing text - The instruction will describe what change to make (e.g., "add that the patient has diabetes", "remove the mention of chest pain", "change the date to January 15th") - Make the requested change while preserving the rest of the text - Ensure the modified text remains grammatically correct and medically accurate 3. SMART_REPORT (task="smart_report"): - Convert the transcription into a structured, professional medical report - Format according to the specified specialty's conventions - Include appropriate sections (e.g., History, Findings, Impression for radiology) - Use proper medical terminology and formatting - If a signature is provided, append it at the end - Ensure the report is ready for clinical use Always respect medical accuracy and never invent or remove medical information. """ # Pydantic models for request bodies class ReviewRequest(BaseModel): text: str specialty: str target: str class ModifyRequest(BaseModel): text: str instruction: str specialty: str target: str class ReportRequest(BaseModel): text: str specialty: str signature: str | None = None def call_ai_task( task: str, specialty: str, input_text: str, instruction: str = None, signature: str = None, ) -> str: """Call OpenAI with the appropriate task and return the output text.""" try: # Build user message based on task if task == "review": user_content = { "task": "review", "specialty": specialty, "input_text": input_text, } elif task == "modify": user_content = { "task": "modify", "specialty": specialty, "input_text": input_text, "instruction": instruction, } elif task == "smart_report": user_content = { "task": "smart_report", "specialty": specialty, "input_text": input_text, "signature": signature, } else: return input_text # Unknown task, return original # Call OpenAI response = client.chat.completions.create( model="gpt-4o-mini", messages=[ {"role": "system", "content": AI_SYSTEM_PROMPT}, {"role": "user", "content": json.dumps(user_content)}, ], temperature=0.3, max_tokens=2000, ) output_text = response.choices[0].message.content.strip() # For smart_report, append signature if provided if task == "smart_report" and signature and signature.strip(): output_text += f"\n\n{signature.strip()}" return output_text except Exception as e: logger.exception("AI task error (%s): %s", task, e) # Return original text on error so UI still works return input_text @app.post("/api/review") async def review(request: ReviewRequest): """Review and clean medical text.""" try: output_text = call_ai_task( task="review", specialty=request.specialty, input_text=request.text, ) return {"output_text": output_text} except Exception as e: logger.exception("Review endpoint error: %s", e) return {"output_text": request.text} # Return original on error @app.post("/api/modify") async def modify(request: ModifyRequest): """Modify text based on voice instruction.""" try: output_text = call_ai_task( task="modify", specialty=request.specialty, input_text=request.text, instruction=request.instruction, ) return {"output_text": output_text} except Exception as e: logger.exception("Modify endpoint error: %s", e) return {"output_text": request.text} # Return original on error @app.post("/api/report") async def generate_report(request: ReportRequest): """Generate structured smart report from transcription.""" try: output_text = call_ai_task( task="smart_report", specialty=request.specialty, input_text=request.text, signature=request.signature, ) return {"output_text": output_text} except Exception as e: logger.exception("Report endpoint error: %s", e) return {"output_text": request.text} # Return original on error