To use the Embeddings API, please go through the detailed information about the API and its parameters from the Documentation.

If you don’t have an API key, please send your request - API Access Request Form.

​
Example

Below is a Python example demonstrating how to perform a question-answering:

​
Function to get similar filings chunks


import requests
import json

def retrieve_similar_filings_chunks(query):
    url = 'https://stockinsights-ai-main-49970eb.d2.zuplo.dev/api/in/v0/documents/embeddings-search'
    headers = {'Content-Type': 'application/json',
              'Authorization': 'Bearer YOUR_API_KEY_HERE'}
    data = {
        "query": query,
        "filters": {
            "types": ["earnings-transcript", "annual-report"],
            'year': '2024'
        }
    }

    response = requests.post(url, headers=headers, data=json.dumps(data))
    return response.json()
    
result = retrieve_similar_filings_chunks("FIIs contribution to Indian Capital Markets")

​
Function to Generate Answers Using OpenAI GPT Model


from openai import OpenAI

# Set your OpenAI API key
client = OpenAI(
  api_key='YOUR_OPEN_AI_KEY_HERE',
)

GPT_MODEL = 'gpt-4o-mini'

# To interact with OpenAI GPT model
def chat_completion_request(messages, model= GPT_MODEL, stream=False, functions=None, function_call=None):
    try:
        response = client.chat.completions.create(
            model=model,
            messages=messages,
            stream = stream,
            functions=functions,
            function_call=function_call
        )
        return response
    except Exception as e:
        print("Unable to generate ChatCompletion response")
        print(f"Exception: {e}")
        return e

# Function to construct knowledge source
def construct_knowledge_source(similar_chunks):
    knowledge_source = ""
    data = similar_chunks.get('data', [])
    
    for entry in data:
        chunk_content = entry.get('text', '')
        company_name = entry.get('company', {}).get('company_name', '')
        document_type = entry.get('document', {}).get('type', '')
        document_link = entry.get('document', {}).get('link', '')
        page_num = entry.get('metadata', {}).get('page_num', '')
        document_link_with_page = f"{document_link}#page={page_num}"
        year = entry.get('document', {}).get('year', '')
        quarter = entry.get('document', {}).get('quarter', '')
        
        # Append the context to each chunk
        knowledge_source += f"This is a section from {document_type} of company {company_name} for FY{year} {quarter if quarter else ''} from {document_link_with_page}: {chunk_content};\n\n"
    
    return knowledge_source

def get_answer_from_public_company_filings(query):
    similar_filings_chunks = retrieve_similar_filings_chunks(query)
    knowledge_source = construct_knowledge_source(similar_filings_chunks)
    response_instructions = """
         Answer user questions based on some parts of the filings/disclosures identified as most relevant to the question.
         Provide a clear and comprehensive response to the question only from the provided context. """

    setup_messages=[
                {'role': 'system', 
                 'content': response_instructions
                },
                {
                'role': 'system',
                'content': knowledge_source
                },
                {'role': 'user',
                'content': query
                }
            ]
    chat_response = chat_completion_request(messages = setup_messages,stream = False)
    message = chat_response.choices[0].message.content
    return message

query = 'What could be the reasons for decreased overseas(FII) contribution in Indian Capital Markets decreasing while the Domestic fund flows contributions are increasing'
get_answer_from_public_company_filings(query)

​
Key Points

  • Replace YOUR_API_KEY_HERE with your actual API key in the Authorization header.
  • Replace YOUR_OPEN_AI_KEY_HERE with your actual OpenAI API key.
  • Modify the query parameter to suit your specific question.
  • Adjust the filters to target specific document types or time periods as needed.