AI Investment Platform (AIIP)

Automated analysis and research, security selection and portfolio construction.

AIIP modules

Innolab is developing the next wave of investment strategies. Our automated strategies are based on artificial intelligence and deep learning. 

AIIP is composed of 4 modules that span the entire investment process of any type of portfolio: from data processing and cleansing, to asset price forecasting, to portfolio construction and risk management.

Each module contains several services and engines that can be used separately or composed into a customized solution.

Partnering with Innolab

We partner with mutual funds and asset managers to enhance existing strategies or develop customized AI investment products.

The AI-powered platform gives a predictive edge on both risk and price. Embedded in the investment process, it can spearhead significant improvements in performance.

Data Ingestion Module (DIM)


Data is the foundation of our entire platform. The Data Ingestion Module is composed of 4 services that fetch, clean, verify and store data from several sources.

Scraping Service

The scraping service continuously scours the internet for price updates. We scrape from sources that allow us to do so and we mainly use the scraped data for verification of higher priority sources.

API Ingestion Service

Our main source of data is RESTFUL API’s to subscription sources. Their servers are polled around the clock for updates.

Pre-Processing Service

This service transforms ingested raw data into standard price series features such as log-returns and covariances.

API Service

We allow access to our own data (predictions) through a RESTFUL API.

Research and Analysis Module (RAM)


Engines under this module extract or condense information from data. Our engines extract non-linear hierarchical features from price series of equities, equity indices, commodities and fixed income.

All 3 engines are based 100% on deep learning with no predefined rules. They are all developed with a strong focus on robustness and generalizability.

Engines are deployed as autonomous systems that work around the clock, analyzing millions of data points each day.

Encoding Engine

The high dimensionality and very high noise level of a price series is a challenge. Our encoding engine significantly reduces the dimensionality of price series and instead represents the series as a low dimensional encoding.

Stock Prediction Engine

The Stock Prediction Engine extracts information from single stocks. Based on historical prices, the probability of a rise in price in the near-term is estimated by an ensemble of robots.

Multi Asset Engine

The Multi Asset Engine extracts information from equity index-, commodity- and fixed income-price series. Based on historical prices, the probability of a rise in the near term is estimated by an ensemble of robots. Each robot is able to forecast any of the asset classes.

Portfolio Construction and Risk Management Module (PM)


The Portfolio Construction and Risk Management module comprises 3 services concerned with selecting and sizing a set of exposures to assets from a pre-defined universe.

Diversification Service

The diversification service selects a small set off assets from a larger ranked list of assets (ranked e.g. by a RAM-service). The selection is a compromise between diversification and rank.

Risk Parity Service

The Risk Parity Service balances the exposure in a given list of assets by risk parity (estimated from a covariance matrix from the DIM-module).

Volatility Planning Service

The Volatility Planning Service adjusts the exposure of a portfolio such that it is a compromise between the estimated risk of the current portfolio and the risk in the near future when positions are planned to be closed.

Trade Execution Module (TEM)


The trade execution module comprises 2 services that allow autonomous trading systems to interact with an exchange to execute orders.

We analyze a broad spectrum of assets around the clock:

Equity indices


Fixed income






Example strategy flow based on AIIP

Noise reduction

Large data sets from multiple sources are pre-processed using a Deep Autoencoder that compresses data and eliminates noise while capturing valuable information.


An ensemble of self-taught deep learning robots, research and analyze worldwide stocks, bonds, commodities and equity/sector indices. The robots are trained to detect multivariate, non-linear relationships in markets.

Portfolio construction

Positions are systematically selected from a ranked asset class lists. Each asset class have a strict predetermined holding period that is optimized through AI and fitted to a strategy. 

Risk management

Risk management are applied to a portfolio with optimizers that considers volatility and correlation between assets. This determine the position sizes and total portfolio risk.

Every capital markets firm should expect to have its future business and economic models challenged by 2022.

Michael Spellacy​



Contact us

Contact information