| from langchain_openai import ChatOpenAI |
| from langchain_core.prompts import ChatPromptTemplate |
| from typing import List, Dict, Any |
| from dotenv import load_dotenv |
| load_dotenv() |
|
|
| def extract_needs(docs, max_concurrency: int = 10) -> List[List[str]]: |
|
|
| llm = ChatOpenAI( |
| model='gpt-4o-mini', |
| temperature=0.2 |
| ) |
|
|
| prompt = ChatPromptTemplate.from_messages( |
| [ |
| ("system", |
| """ |
| You are a data extraction assistant helping a grant review team evaluate funding applications from schools. Your task is to extract a structured list of specific items or services that the school is requesting funding for. |
| |
| ## Instructions: |
| |
| - Users will submit excerpts from grant application letters. |
| - From each letter, extract only the tangible items or clearly defined services the school wants to use the grant for. |
| - Output the extracted items as a **clean, comma-separated list**, with no additional explanation or formatting. |
| - Do not include abstract goals or general program names (e.g., "Arts Award program" or "student development"). |
| - Focus on concrete nouns that represent resources or services the grant would directly fund (e.g., "paint", "laptops", "counseling sessions", "sports equipment"). |
| |
| ## Example: |
| |
| **User Input:** |
| _I work in an alternative provision supporting disadvantaged children with SEND. Many of our students face significant challenges in communication and emotional expression. |
| Art is a powerful tool that allows them to process feelings, build confidence, and develop essential life skills. |
| With your support, we would purchase paints, canvases, clay, and sketchbooks to run the Arts Award program. |
| This would give our students a creative voice and a sense of achievement. Additionally, we could bring in an art therapist weekly, providing vital emotional support. |
| Your funding would transform lives, giving these children hope and opportunity._ |
| |
| **Your Output:** |
| paints, canvases, clay, sketchbooks, art therapist |
| """ |
| ), |
| ("human", "{input}") |
| ] |
| ) |
|
|
| chain = prompt | llm |
|
|
| responses = chain.batch(docs, config={"max_concurrency": max_concurrency}) |
| |
| result = [] |
|
|
| for response in responses: |
| |
| content = response.content |
| |
| items = [item.strip() for item in content.split(',')] |
| |
| result.append(items) |
| |
| return result |
|
|
