updated schemas and drafting wip

Author: sarosit

jao-backend-schemas/src/jao_backend_schemas/draft.py

@@ -0,0 +1,11 @@
+from pydantic import BaseModel
+from .vacancies import VacancyListing
+
+
+class DraftRequest(BaseModel):
+    description: str
+    similar_vacancies: list[VacancyListing]
+
+
+class DraftResponse(BaseModel):
+    draft: str

jao-backend/src/jao_backend/api/v0/endpoints.py

@@ -10,6 +10,8 @@
 
 from jao_backend_schemas.advice import AdviceResponse
 from jao_backend_schemas.advice import AdviceRequest
+from jao_backend_schemas.draft import DraftResponse
+from jao_backend_schemas.draft import DraftRequest
 from jao_backend_schemas.maps import AreaFrequenciesResponse
 from jao_backend_schemas.plots import PlotlyFiguresResponse
 from jao_backend_schemas.vacancies import SimilarVacanciesResponse
@@ -19,8 +21,8 @@
 from jao_backend.common.text_processing.clean_oleeo import parse_oleeo_bbcode
 from jao_backend.embeddings.models import EmbeddingTag
 from jao_backend.vacancies.models import VacancyEmbedding
-from jao_backend.advice.advice import AdviceService
-advice_service = AdviceService()
+from jao_backend.llm.llm_service import LLMService
+llm_service = LLMService()
 
 logger = logging.getLogger(__name__)
 
@@ -78,6 +80,23 @@ def get_similar_vacancies(text, top_n=10):
     ]
 
 
+@api.post("/similar_adverts", response=SimilarVacanciesResponse)
+def similar_adverts(
+    request: HttpRequest, payload: JobDescriptionRequest
+) -> SimilarVacanciesResponse:
+    similar_vacancies_list = [
+        VacancyListing.model_validate(
+            {
+                "job_title": vacancy.title,
+                "full_job_desc": parse_oleeo_bbcode(vacancy.description),
+                "vacancy_id": vacancy.pk,
+            }
+        )
+        for vacancy in get_similar_vacancies_cached(payload.description, top_n=10)
+    ]
+    return SimilarVacanciesResponse(similar_vacancies=similar_vacancies_list)
+
+
 @api.post("/advice")
 def advice(request: HttpRequest, payload: AdviceRequest) -> StreamingHttpResponse:
     formatted_vacancies = [
@@ -88,9 +107,9 @@ def advice(request: HttpRequest, payload: AdviceRequest) -> StreamingHttpRespons
 
     def generate():
         """Generator for streaming response"""
-        for chunk in advice_service.get_advice(payload.description,
-                                               formatted_vacancies,
-                                               payload.advice_type):
+        for chunk in llm_service.get_advice(payload.description,
+                                            formatted_vacancies,
+                                            payload.advice_type):
             logger.debug(f"Streaming chunk: {chunk}")
             yield f"data: {json.dumps({'content': chunk})}\n\n"
 
@@ -101,69 +120,71 @@ def generate():
     )
 
 
-@api.post("/similar_adverts", response=SimilarVacanciesResponse)
-def similar_adverts(
-    request: HttpRequest, payload: JobDescriptionRequest
-) -> SimilarVacanciesResponse:
-    similar_vacancies_list = [
-        VacancyListing.model_validate(
-            {
-                "job_title": vacancy.title,
-                "full_job_desc": parse_oleeo_bbcode(vacancy.description),
-                "vacancy_id": vacancy.pk,
-            }
-        )
-        for vacancy in get_similar_vacancies_cached(payload.description, top_n=10)
+@api.post("/draft")
+def draft(request: HttpRequest, payload: DraftResponse) -> StreamingHttpResponse:
+    formatted_vacancies = [
+        f"Job Title: {vacancy.job_title}\nDescription: {
+            parse_oleeo_bbcode(vacancy.full_job_desc)}"
+        for vacancy in payload.similar_vacancies
     ]
-    return SimilarVacanciesResponse(similar_vacancies=similar_vacancies_list)
-
-# Write a query using the similar vacancies data from application stastics -
-# aggregated application statistic model
-
 
-@api.post("/similar_advert_plots")
-def similar_advert_plots(
-    request: HttpRequest, payload: JobDescriptionRequest
-) -> PlotlyFiguresResponse:
-    """
-    Get a graph of the job description.
-    """
-    # Stub: this required aggregated data
-    logger.info(
-        "STUB: similar_advert_plots endpoint called with description: %s",
-        payload.description,
+    def generate():
+        for chunk in llm_service.get_draft(payload.description, formatted_vacancies):
+            yield f"data: {json.dumps({'content': chunk})}\n\n"
+    yield "data: [DONE]\n\n"
+    return StreamingHttpResponse(
+        generate(),
+        content_type='text/event-stream'
     )
-    graphs = []
-    return PlotlyFiguresResponse(plotly_figures=graphs)
-
-# ADD Skills Ingester to write skills to DB
 
 
-@api.post("/skills_plots")
-def skills_plots(request, payload: JobDescriptionRequest) -> PlotlyFiguresResponse:
-    """
-    Get a graph of the job description.
-    """
-    # Stub: skills are not ingested right now.
-    logger.info(
-        "STUB: skills_plots endpoint called with description: %s", payload.description
-    )
-    graphs = []
-    result = PlotlyFiguresResponse(plotly_figures=graphs)
-    return result
-
-# Maybe REMOVE the location response
+# Write a query using the similar vacancies data from application stastics -
+# aggregated application statistic model
 
 
-@api.post("/applicant_locations")
-def applicant_locations(
-    request, payload: JobDescriptionRequest
-) -> AreaFrequenciesResponse:
-    # Stub: OLEEO ingestion of locations is TBD
-    logger.info(
-        "STUB: applicant_locations endpoint called with description: %s",
-        payload.description,
-    )
-    area_frequencies: AreaFrequencyProperties = []
-    # TODO: populate area_frequencies with instances of AreaFrequencyProperties from the database.
-    return AreaFrequenciesResponse(area_frequencies=area_frequencies)
+# @api.post("/similar_advert_plots")
+# def similar_advert_plots(
+#     request: HttpRequest, payload: JobDescriptionRequest
+# ) -> PlotlyFiguresResponse:
+#     """
+#     Get a graph of the job description.
+#     """
+#     # Stub: this required aggregated data
+#     logger.info(
+#         "STUB: similar_advert_plots endpoint called with description: %s",
+#         payload.description,
+#     )
+#     graphs = []
+#     return PlotlyFiguresResponse(plotly_figures=graphs)
+#
+# # ADD Skills Ingester to write skills to DB
+#
+#
+# @api.post("/skills_plots")
+# def skills_plots(request, payload: JobDescriptionRequest) -> PlotlyFiguresResponse:
+#     """
+#     Get a graph of the job description.
+#     """
+#     # Stub: skills are not ingested right now.
+#     logger.info(
+#         "STUB: skills_plots endpoint called with description: %s", payload.description
+#     )
+#     graphs = []
+#     result = PlotlyFiguresResponse(plotly_figures=graphs)
+#     return result
+#
+# # Maybe REMOVE the location response
+#
+#
+# @api.post("/applicant_locations")
+# def applicant_locations(
+#     request, payload: JobDescriptionRequest
+# ) -> AreaFrequenciesResponse:
+#     # Stub: OLEEO ingestion of locations is TBD
+#     logger.info(
+#         "STUB: applicant_locations endpoint called with description: %s",
+#         payload.description,
+#     )
+#     area_frequencies: AreaFrequencyProperties = []
+#     # TODO: populate area_frequencies with instances of AreaFrequencyProperties from the database.
+#     return AreaFrequenciesResponse(area_frequencies=area_frequencies)

…ckend/src/jao_backend/advice/__init__.py …-backend/src/jao_backend/llm/__init__.pyjao-backend/src/jao_backend/advice/__init__.py renamed to jao-backend/src/jao_backend/llm/__init__.py jao-backend/src/jao_backend/advice/__init__.py renamed to jao-backend/src/jao_backend/llm/__init__.py

@@ -3,4 +3,4 @@
 
 class AdviceConfig(AppConfig):
     default_auto_field = 'django.db.models.BigAutoField'
-    name = 'jao_backend.advice'
+    name = 'jao_backend.llm'

…-backend/src/jao_backend/advice/admin.py jao-backend/src/jao_backend/llm/admin.pyjao-backend/src/jao_backend/advice/admin.py renamed to jao-backend/src/jao_backend/llm/admin.py jao-backend/src/jao_backend/advice/admin.py renamed to jao-backend/src/jao_backend/llm/admin.py

@@ -10,8 +10,9 @@
 from jao_backend.application_statistics.models.lists import Gender, Disability
 from jao_backend.application_statistics.models.statistics import AggregatedApplicationStatistic
 from jao_backend_schemas.advice import AdviceResponse
+from jao_backend_schemas.draft import DraftResponse
 from jao_backend.settings.common import EMBEDDING_TAG_JOB_TITLE_RESPONSIBILITIES_ID
-from jao_backend.advice.prompts import ADVICE_PROMPTS
+from jao_backend.llm.prompts import ADVICE_PROMPTS, DRAFT_PROMPT
 
 LITELLM_API_BASE = settings.LITELLM_API_BASE
 LITELLM_CUSTOM_PROVIDER = settings.LITELLM_CUSTOM_PROVIDER
@@ -20,7 +21,7 @@
 logger = logging.getLogger(__name__)
 
 
-class AdviceService():
+class LLMService():
 
     def __init__(self):
         self.tag = EmbeddingTag.get_tag(
@@ -79,30 +80,45 @@ def _update_rag_content(self, user_input, filters):
                 'No similar vacancies found with the applied filters.'))
         return updated_vacancies
 
+    def _draft_handler(self, user_input, rag_content):
+
+        prompt_config = DRAFT_PROMPT
+
+        messages = [
+            {"role": "system", "content": prompt_config["system"]},
+            {"role": "user", "content": prompt_config["user_template"].format(
+                rag_content=rag_content,
+                user_input=user_input
+            )}
+        ]
+
+        try:
+            return completion(
+                model=self.model,
+                messages=messages,
+                stream=True,
+                max_tokens=1500,
+                api_base=LITELLM_API_BASE,
+                custom_llm_provider=LITELLM_CUSTOM_PROVIDER
+            )
+        except Exception as e:
+            logger.error(f'Error generating advice: {str(e)}')
+            raise
+
     def _advice_handler(self, user_input, rag_content, advice_type, options=None):
-        # For non-general advice, apply filters and get updated RAG content
         if advice_type != "general" and options:
             filters = self.build_filters(options)
             try:
                 updated_vacancies = self._update_rag_content(
                     user_input, filters)
-                # Rebuild RAG content with filtered vacancies
                 rag_content = "\n\n".join(
                     [f"Job Ad {i+1}:\n{vacancy.vacancy.full_job_desc}"
                      for i, vacancy in enumerate(updated_vacancies)]
                 )
             except Exception as e:
                 logger.error(f'Error filtering vacancies: {str(e)}')
-                # Fall back to original rag_content if filtering fails
 
         prompt_config = ADVICE_PROMPTS.get(advice_type)
-        print("********")
-        print("********")
-        print(advice_type)
-        print(prompt_config)
-
-        print("********")
-        print("********")
         if not prompt_config:
             raise ValueError(f"Unknown advice type: {advice_type}")
 
@@ -148,3 +164,24 @@ def get_advice(self, user_input, similar_vacancies, advice_type,
         except Exception as e:
             logger.error(f"Error generating advice with LiteLLM: {str(e)}")
             yield "Sorry, I'm unable to generate advice at the moment. Please try again later."
+
+    def get_draft(self, user_input, similar_vacancies):
+        rag_content = "\n\n".join(
+            [f"Job Ad {i+1}:\n{ad}" for i, ad in enumerate(similar_vacancies)])
+        try:
+            response = self._draft_handler(
+                user_input, rag_content)
+            for chunk in response:
+                if chunk.choices[0].delta.content:
+                    yield chunk.choices[0].delta.content
+
+        except APIConnectionError as e:
+            logger.error(
+                "Connection refused to the completion service. "
+                "Ensure the service is running and accessible: %s",
+                e,
+            )
+            raise
+        except Exception as e:
+            logger.error(f"Error generating advice with LiteLLM: {str(e)}")
+            yield "Sorry, I'm unable to generate drafts at the moment. Please try again later."