Aichemy

🚧AI Hooks for your Database🚧

Aichemy is still in development and everything in this readme is a possible concept yet.

Aichemy makes it easy to use AI in your database. Consider this as an observation layer that watch the changes in DB and act when possible. With powerful AI, many seemingly complex update logics can be implemented:

• When age is updated, combining with title to guess interested(bool, if user wants to buy our services).
• When bio is updated, combing with name to generate a personalized sale_dm(string, content of DM)
• ...

All you need to do is focusing on the business logic, and write the instructions clearly.

Install

pip install aichemy-py SQLAlchemy

aichemy only supports SQLAlchemy for now.

QuickStart

1. Design your DB

from sqlalchemy import create_engine, Column, Integer, String
from sqlalchemy.orm import declarative_base, sessionmaker


class User(Base):
    __tablename__ = "users"
    id = Column(Integer, primary_key=True)
    name = Column(String)
    title = Column(String)
    intro = Column(String)
    dm_message = Column(String, nullable=True)
    
    
engine = create_engine("sqlite:///test.db")
Base.metadata.create_all(engine)
Session = sessionmaker(bind=engine)

2. Write aichemy functions

import aichemy as ac

@ac.watch(User.intro).with(User.name, User.title).to(User.dm_message)
def write_dm(row: dict, results: dict):
  """Based on the user information, write a thank-you DM for our AI product."""  
  # 1. pre-process
  row[User.name] = row[User.name].capitalize()
	
  # 2. pass your descriptions
  yield {
    User.intro: "The self-introduction of the user",
    User.dm_message: "Write the dm_messages"
  }
  
  # 3. post-process
  results[User.dm_message] = results[User.dm_message][:200]

• row is a dict containing the values of (User.intro, User.name, User.title)
• You can use docstring of the function to write your instructions
• Before yield, you can modify the values inside row for pre-process. At this time, results is always empty and any changes to results is meaningless.
• You can yield the descriptions for both input and output, the descriptions will be used in the process of AI.
• After yield, you can modify the return values inside results before they're committed to Database

3. Start watching

write_dm.start()

with Session() as session:
    user = User(name="John", title="Software Engineer", intro="I love AI and basketball")
    session.add(user)
    session.commit()
    
with Session() as session:
    user = session.query(User).all()
    print([u.dm_message for u in user])

By default, aichemy works in sync mode. That means the watched fields' writing will be blocked until the aichemy function finish. You can enable nowait mode to stop aichemy block any of your DB operations.

@ac.watch(User.intro).with(User.name, User.title).to(User.dm_message).nowait()
def write_dm(row: dict, results: dict):
  ...

4. Debuging

Debug the prompt is easy in aichemy:

write_dm.debug({
  User.name: "Gus",
  User.title: "Founder of Memobase",
  User.intro: "I'm a programmer",
})