AIDE from weco serves as an AI agent for Machine Learning. Its primary capability involves designing and optimizing Machine Learning pipelines based on user instructions. AIDE has been developed to handle both complex business problems and academic research. Users can instruct AIDE in natural language, making it accessible to both experts, who can provide detailed instructions, and novices, who can let AIDE guide the process. A key feature of AIDE is its iterative process of analysing data, crafting, evaluating, and refining solutions. This approach is driven by Large Language Models (LLMs) that enable the system to write code and systematically seek out improved designs. The solution AIDE produces can then be applied in relevant tasks, distinguishing it from traditional AutoML solutions and data science consultancy services.
Autonomous Code Optimization: Weco utilizes evaluation-driven code optimization, allowing users to provide source code and an evaluation script for iterative performance improvements.
Observable Search Tree: Users can explore a tree structure where each node displays code, terminal output, and performance metrics, enhancing transparency in the optimization process.
Steerable Mid-Run Adjustments: Users can introduce new ideas or papers during the optimization process, creating new branches in the search for better solutions.
Model Neutrality: The platform supports various models, including GPT, Claude, and others, allowing users to bring their own API keys for flexibility.
Diverse Use Cases: Weco addresses a wide range of applications, including fraud detection, prompt engineering, inference optimization, and hyperparameter tuning, demonstrating its versatility across different domains.
Fraud Detection: Optimizes end-to-end transaction-fraud machine learning pipelines, improving AUC from 0.91 to 0.93.
Prompt Engineering: Enhances accuracy in vision-language tasks, achieving a 32.4% increase in chart-to-CSV extraction accuracy.
Inference Optimization: Increases speed by 47.7% for causal self-attention in GPT-style language models.
Hyperparameter Optimization: Reduces RMSLE by 91.3% for molecular property prediction in materials science.
Route Planning: Optimizes urban transportation routes, achieving a score improvement from zero to 34.0 million under congestion constraints.
Enables recursive self-improvement of AI models, enhancing their performance over time without manual intervention.
Provides a platform for optimizing machine learning pipelines, resulting in measurable improvements in metrics such as accuracy, speed, and cost.
Facilitates the integration of human judgment into AI processes, ensuring that AI systems remain aligned with human values and decision-making.
Offers a collaborative environment where AI can publish its findings, allowing human researchers to build upon AI-generated insights and innovations.
Supports a wide range of applications, from fraud detection to logistics optimization, demonstrating versatility across various industries and use cases.
Free Plan: 20 free credits, approximately 100 steps of Weco-hosted autoresearch, and free access to Weco Observe.
Bring Your Own Keys (BYOK): Use your OpenAI, Anthropic, or Gemini keys with up to 10,000 experiments per month.
Pay As You Go: Includes everything in the Free plan, plus access to all models and priority support.
