changed prompt and fixed typo

app.py

@@ -33,7 +33,7 @@ def call_endpoint(resume_preprocessed):
     return response_data
 
 
-def postprocess_vancy(vacancies, resume):
+def postprocess_vacancy(vacancies, resume):
     if "prediction" in vacancies:
         prediction = vacancies["prediction"]
         if isinstance(prediction, list):
@@ -75,10 +75,10 @@ def postprocess_vancy(vacancies, resume):
                     formatted_vacancy = vacancy.replace(".,", "<br/>")
                     formatted_vacancy = f"VACATURE {i + 1}:<br/>{formatted_vacancy}"
                     matches_html = "<br/> - ".join(matched_skills)
-                    macthes_formatted = (
+                    matches_formatted = (
                         f"Score: {len(matched_skills)}<br/>-{matches_html}"
                     )
-                    html_table += f"<tr><td>{formatted_vacancy}</td><td>{macthes_formatted}</td></tr>"
+                    html_table += f"<tr><td>{formatted_vacancy}</td><td>{matches_formatted}</td></tr>"
             html_table += "</table>"
             return html_table
 
@@ -89,7 +89,7 @@ def search(resume):
     original_resume = resume
     resume_preprocessed = preprocess_resume(llm, resume)
     vacancies = call_endpoint(resume_preprocessed)
-    vacancies_formatted = postprocess_vancy(vacancies, original_resume)
+    vacancies_formatted = postprocess_vacancy(vacancies, original_resume)
     return vacancies_formatted
 
 

prompts/matches.py

@@ -112,11 +112,19 @@ def get_skills_match(llm, vacancy, resume) -> SequentialChain:
       
       ```
       RESUME:
+    
       {resume}
       ```
       
-      Both the vacancy and resume are delimited by three backticks. 
-      Can you list any matches you find in both the vacancy and the resume. 
+      Given the vacancy and resume above, both delimited by three backticks, fill in the following JSON object below.
+      The goal is to count matches between the vacancy and the resume and the number of occurrences.
+      For the content, take into the account the following rules:
+      - the final json object should not have duplicates
+      - words with similar meaning should be counted as one and the same, for example 'magazijn' and 'magazijnier' should just be magazijn
+      - words that hold no significant meaning should not be listed, such as 'ervaring' on its own
+      - take into account the meaning of the matches and whether or not it is relevant
+      - only take into account matches that are related to job specific content, experience and/or location
+      - only take the key words that are important to the specific job, limit it to the 10 most relevant matches at most
       
       Each match is constructed as the following JSON object: