Filecoin ML Engine

Filecoin ML engine is an onchain ML agent built to provide data collection, extraction, and computation on top of the Filecoin virtual machine. The major components of the Filecoin ML engine include -

1. Data lake where the users contribute data and get incentivized onchain
2. ML agent contracts - ML smart contracts directly pluggable into the onchain Data lake and are trained on top of the Data lake schemas
3. Prediction engine - Once the ML models are trained they can be directly used for predictions on top of onchain and off-chain data
4. LLM inference engine - this engine is built on top of Lilypad to draw general inferences from the data.

All these components combined create an end-to-end ML engine where the users can build ML agents dapps on top of it.

This project utilizes FVM runtime to create on-chain and off-chain data aggregation where the multiple dapps can publish data can compute on top of it. There are 3 major smart contracts that power the ML engine -

1. Data layer smart contract where the data is arranged in the form of schemas, users can create multiple schemas and get incentivized to contribute data to those schemas.
2. ML Model smart contracts - these are ML agents where we can directly plugin the Data layer smart contract for data ingestion and model training. We have created 4 ML agents - KNN agent, Anomaly detection agent, Linear regression, and logistic regression agent.
3. LLM Oracle smart contract - this smart contract calls the LLM API with the data through Lilypad in order to derive general inference from the LLM model.