Private markets like Venture Capital, Real Estate Commercial Pooled Investments & the likes are a growing area of interest for retail investors recently. Driven by the need to invest in pre-IPO firms with aggressive growth rates while accepting the risks that come along with it. Additionally, an opportunity to diversify a portfolio at a fraction of the cost vs investing at high dollar amounts in VC led rounds.

One of the firms that facilitates such a platform for accredited retail investors to buy into a portfolio of early to late-stage pre-IPO firms alongwith Commercial Real Estate Pooled Investments has been struggling to provide a more catered user experience for its customers in how to present the data so investors can make informed investment decisions. This led to a partnership where we are looking at various AI capabilities to solve this multifold problem. 

Solution Needs:

Firm is looking to target 2 specific areas to ensure stellar client experience: 

1. Providing clients with an interface that generates the content that client is looking for vs providing detailed fund documentation or company financial reports for them to search & selectively understand.

2. Understanding the client profile and building a matching algorithm that provides clients with investment options that are specifically tailored to their needs and objectives.

While reviewing the problem statement and outcomes, we came at the following step by step approach towards a solution:

1. Understanding & segmentation of the client profiles:

a. Consolidating all client profile data into one exhaustive data store alongwith historical investment mapping

b. Segmentation of the client Data to various client personas with respective investment interests and needs c. Testing new investments against these existing personas to ensure no mapping mismatch

2. Investment Information data lake & Persona Match:

a. Bringing all fund & venture pooled investments data into a data lake for ease of access

b. Segmenting the client persona match to investments based on historical transactions 

3. Building out Recommender System to provide personalized matching to client investment needs:

a. Utilizing Unsupervised ML algorithms like KNN & advanced boosters like XGBoost to build predictive models for persona to investment match

b. Through over-sampling methods building out comprehensive mapping data against base model

c. Creating a run-time Deep Neural Net. to process client’s historical investment patterns & predict fund match that is displayed to the client

4. Building out a Generative AI Investment Research module for fast & specific data access

a. Utilizing Mistral 8x7B open source base LLM to bring generative context into investment research

b. Fine Tuning LLM alongside Investment Data Lake to train a firm centric LLM

c. Response Cache provides ability to reduce computational intensity for future prompts

While the initial implementation does require at least 2-3 years of data, the firm is already seeing high accuracy in investment matches that clients have purchased in the recent past. Queries on investment data research is continuing to improve due to our fine-tuning approach and making cache persistent for future prompt.

This is one of the many ways Discovery Partners is helping its clients utilize Deep Learning and Generative AI to innovate and produce results without significant human capital investment. Reach out to us for more information!