tech_stuff.py

import os 

api_key = os.environ.get("api_key")

prefix = "Following the Entrepreneur First Edge compatibility and intersection approach, with the potential of building a billion dollar company. \n\nMINIMUM POSSIBLE SKILL AND PROFILE OVERLAP! HIGHLY COMPLEMENTARY PROFILES TO HELP THIS PERSON CO-FOUND A BILLION-DOLLARS COMPANY\n"

how_does_it_work = """

This app helps you find a potential co-founder using AI and LLMs. 

#### Co-finder: 

The main component is the OpenAI embedding retrieval model that find similarities based on the information given in the cohort's Dashboard

Link to documentation: https://platform.openai.com/docs/guides/embeddings

Because this is a similarity engine and we are looking for complementary profiles we add a small prefix before the profile we want to match.

Example:

Profile A is an expert in biotech and is looking for someone business oriented

The new augmented version of this profile will be 

Prefix + Profile description

The intuition is to steer the retrieval to make sure it consider complementary profiles as being the most similar to the augmented profile version.

You can change the prefix directly in this page.

Note, the finder works way better using ColbertV2 embeddings but it was a nightmare to deploy on huggingface, and since it is more of an experiment and either way nothing will replace human interactions, openai embeddings should be enough for now.

### Simulator:

This is simply a call to GPT4 with the two profiles descriptions. You can adapt the system prompt here too.


"""


system_prompt = """
You are an expert in assessing startup co-founding teams and finding their potential to build billion-dollar companies. You use the Entrepreneur First Edge intersection method to propose (Strictly respect the Markdown format):

- ### How those Edges might intersect
<concise bullet points>

- ### What kind of potential ideas/industries should be discussed and why they might be hair on fire problems
< minimum 10 concise bullet points>

- ### What common belief could be leveraged
<concise bullet points>
"""