jacob-mathison/doczyai-pipelines
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Activity

Fix prompt to ensure list-of-dicts-of-str

Browse Source
This commit is contained in:
Katon Minhas
2026-02-04 12:04:48 -05:00
parent 16ead7d5f3
commit 7ddd9dc192
1 changed files with 14 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/prompts/prompt_templates.py
+14 -6
View File
@@ -1442,17 +1442,20 @@ Examples:
Briefly explain your reasoning. After your explanation, include the heading "FINAL PROVIDER INFO:" followed by a properly formatted JSON array with your final output.
[EXTRACTION AND FORMATTING INSTRUCTIONS]
[EXTRACTION INSTRUCTIONS]
- First, analyze the text thoroughly and identify all provider entities
- Return TIN and NPIs that are found on the page even if they aren't explicitly labeled as such.
- If the table contains multiple name columns (e.g. 'System Name' and 'Facility Name'), YOU MUST extract the specific 'Facility Name' or 'Hospital Name' column.
- IMPORTANT VALIDATION REQUIREMENTS:
* TIN must be exactly 9 digits (no more, no less)
* NPI must be exactly 10 digits (no more, no less)
* Double-check the digit count for each TIN and NPI before including them in the response
- In your response, include ONLY ONE SINGLE properly formatted JSON array, after your explanation
- Each provider must be an object within this array
* CAREFULLY double-check the digit count for each TIN and NPI before including them in the response
[FORMATTING INSTRUCTIONS]
- In your response, include ONLY ONE SINGLE properly formatted JSON List-of-Dicts-of-Str, after your explanation
- Each provider must be an object (DICT) within this array
- All provider objects must have these exact keys: "TIN", "NPI", "NAME"
- ALl dict values must be strings - if there are multiple different TINs or NPIs for an identical NAME, put them in different dictionaries.
- Use "N/A" for any missing values
- A provider record is considered valid as long as either a TIN, NPI, or NAME is present. If any one or two of those fields are missing, use "N/A" for those fields.
- Your final JSON array must begin with an opening bracket '[' and end with a closing bracket ']'
@@ -1467,13 +1470,18 @@ FINAL PROVIDER INFO:
"NPI": "1234567890",
"NAME": "Main Provider Group"
},
{
"TIN": "124681012",
"NPI": "N/A",
"NAME": "Main Provider Group"
},
{
"TIN": "987654321",
"NPI": null,
"NPI": "N/A,
"NAME": "Dr. John Smith",
},
{
"TIN": null,
"TIN": "N/A",
"NPI": "1234567890",
"NAME": "Jane Doe, CRNA"
}