FlashLearn

Integrate agents into your enterprise ETL workflows in minutes.

Agent in 3 steps

Understand and reporduce each step in the pipeline - everytime!

1. Learn Skills

Learn to perform tasks by just providing description.
from flashlearn.skills.learn_skill import LearnSkill
from openai import OpenAI
from flashlearn.skills.general_skill import GeneralSkill

# Instantiate your pipeline "estimator" or "transformer" learner
learner = LearnSkill(model_name="gpt-4o-mini", client=OpenAI())

# Define your custom task instructions and sample data (if available)
skill = learner.learn_skill(
    df=[],  # You can pass an initial dataset here if needed
    task=(
        "Evaluate how likely the user is to buy my product based on the sentiment in their comment, "
        "return an integer 1-100 on key 'likely_to_buy', and a short explanation on key 'reason'."
    ),
)

# Save the learned skill for future use
skill.save("evaluate_buy_comments_skill.json")

2. Use Skills anywhere

Skills can be stored and used from any system downstream.
# Example input: list of dictionaries representing user comments
user_inputs = [
    {"comment_text": "I love this product, it's everything I wanted!"},
    {"comment_text": "Not impressed... wouldn't consider buying this."},
]

# To use the saved skill, load it and run the tasks in parallel
with open("evaluate_buy_comments_skill.json", "r", encoding="utf-8") as file:
    definition = json.load(file) 
    
# Load the previously saved skill
skill = GeneralSkill.load_skill(definition)

# Create tasks from the list of user inputs
tasks = skill.create_tasks(user_inputs)

# Run tasks in parallel and get structured results
results = skill.run_tasks_in_parallel(tasks)

print("Evaluation Results:")
print(results)

3. Next steps

You always get structured results that can be easy used in next steps.
# Suppose 'flash_results' is the dictionary with structured LLM outputs
for idx, result in flash_results.items():
    desired_score = result["likely_to_buy"]
    reason_text = result["reason"]
    # Now do something with the score and reason, e.g., store in DB or pass to next step
    print(f"Comment #{idx} => Score: {desired_score}, Reason: {reason_text}")

Developers love it!

Interesting, an agent library that doesn't feel like wrestling with an octopus. The fit/predict pattern is cool, intuitive for anyone with a data science background.
Brilliant-Day2748
Most agent frameworks feel way too heavy, so a minimal, JSON-driven approach is a game-changer.
Dan27138
I'm just starting to try to understand how to use agents and this looks like a perfect way to start playing around with stuff!
plyr500

Why FlashLearn?

Build fast, predictable and mantainable agent infrastucture with ease.

Easy Integration

Modular APIs that smoothly connect with your existing enterprise systems.

Structured JSON

Provides uniform JSON outputs for seamless downstream integration

High Throughput

Handle 1,000 calls per second for large-scale operations.

Multi-LLM Support

Works with on-prem and popular providers like OpenAI, LiteLLM, and Ollama

Adam Giacomelli
Gal Giacomelli

Who are we?

We are a seasoned team of developers from Slovenia with over 15 years of AI and big data experience, focusing on making AI development as simple as possible.

Managed service

Book your time with founders and develop world class solutions!

Simplify your AI journey