update: file app

app.py

@@ -6,7 +6,7 @@ from fastapi.middleware.cors import CORSMiddleware
 import cv2
 import numpy as np
 from src.config.llm import llm
-from src.prompt.promt import format_prompt
+from src.prompt.promt import format_prompt, matching_jd_prompt
 from langchain_core.output_parsers import JsonOutputParser
 import uvicorn
 from io import BytesIO
@@ -18,6 +18,8 @@ import os
 import functools
 import threading
 from src.inference.segment_inference import inference
+from PIL import Image
+
 load_dotenv()
 app = FastAPI(docs_url="/")
 app.add_middleware(
@@ -28,6 +30,7 @@ app.add_middleware(
     allow_headers=["*"],
 )
 executor = ThreadPoolExecutor(max_workers=int(os.cpu_count() + 4))
+parser = JsonOutputParser()
 
 
 def run_in_thread(func, *args, **kwargs):
@@ -43,7 +46,6 @@ def run_in_thread(func, *args, **kwargs):
 
 
 def predict_func(threshold_confidence, threshold_iou, image):
-
     image = np.frombuffer(image, np.uint8)
     image = cv2.imdecode(image, cv2.IMREAD_COLOR)
     outputs = inference(
@@ -52,11 +54,19 @@ def predict_func(threshold_confidence, threshold_iou, image):
         threshold_iou=threshold_iou,
     )
     text = extract_text(outputs=outputs, image_origin=image)
-    image = draw_bounding_boxes(image, outputs)
+    image_with_boxes = draw_bounding_boxes(image, outputs)
+    if isinstance(image_with_boxes, np.ndarray):
+        image_rgb = cv2.cvtColor(image_with_boxes, cv2.COLOR_BGR2RGB)
+        image_pil = Image.fromarray(image_rgb)
+    elif isinstance(image_with_boxes, Image.Image):
+        image_pil = image_with_boxes
+    else:
+        raise TypeError(f"Unsupported image type: {type(image_with_boxes)}")
+
+    # Encode image to base64
     buffer = BytesIO()
-    image.save(buffer, format="JPEG")
+    image_pil.save(buffer, format="JPEG")
     buffer.seek(0)
-
     image_base64 = base64.b64encode(buffer.getvalue()).decode("utf-8")
     response = {"outputs": text, "image_base64": image_base64}
     return response
@@ -82,22 +92,49 @@ async def predict(
 
 class LLMRequest(BaseModel):
     text: str = Field(..., title="Text to generate completion")
+    job_desciption: str = Field(
+        default=None, title="Job Description to match with resume"
+    )
+
+
+def reformat_fn(data):
+    chain = format_prompt | llm | parser
+    response = chain.invoke({"user_input": data})
+    return response
+
+
+@app.post("/reformat_output", status_code=status.HTTP_200_OK)
+async def reformat_output(data: LLMRequest):
+    try:
+        response = await run_in_thread(reformat_fn, data.text)
+        return JSONResponse(content=response, status_code=status.HTTP_200_OK)
+    except Exception as e:
+        response = {"error": str(e)}
+        return JSONResponse(content=response, status_code=status.HTTP_400_BAD_REQUEST)
 
 
-def call_llm(data):
-    input = format_prompt.format(input=data)
-    response = llm.invoke(input)
-    response = JsonOutputParser().parse(response)
+def matching_job_desciption_fn(data: LLMRequest):
+    job_description = data.job_desciption
+    resume_input = data.text
+    chain = matching_jd_prompt | llm | parser
+    response = chain.invoke(
+        {"job_description": job_description, "resume_input": resume_input}
+    )
+    print(response)
     return response
 
 
-@app.post("/llm", status_code=status.HTTP_200_OK)
-async def llm_predict(data: LLMRequest):
+@app.post("/matching_job_desciption", status_code=status.HTTP_200_OK)
+async def matching_job_desciption(data: LLMRequest):
+    if data.job_desciption is None:
+        response = {"error": "Job Description is required"}
+        return JSONResponse(content=response, status_code=status.HTTP_400_BAD_REQUEST)
     try:
-        response = await run_in_thread(call_llm, data.text)
+        response = await run_in_thread(matching_job_desciption_fn, data)
         return JSONResponse(content=response, status_code=status.HTTP_200_OK)
     except Exception as e:
         response = {"error": str(e)}
+        print(response)
         return JSONResponse(content=response, status_code=status.HTTP_400_BAD_REQUEST)