Regulated investment platform
Investors get matched deals and document-backed answers, while every decision stays theirs
For a live private investment marketplace, we built the investor side of its AI: deal recommendations, a summary that checks the issuer's, an assistant scoped to one deal and a drafted expression of interest. All of it is read-only, and people make every call.
Financial ServicesPrivate investment marketplaceAI added to a live product
The challenge
Investors read long, inconsistent deal documents on their own and issuers found potential investors by hand, while many users were older and wanted people firmly in charge.
What we did
We built read-only investor agents as APIs: deal recommendations, a market check on each deal, a deal-scoped assistant, a drafted expression of interest and investor matching for issuers.
The outcome
The client took the delivered recommendation, investor relations and investor assistant agents into integration and acceptance testing.
- Recommends
- Deals from interests, past activity and risk profile
- Assistant
- Answers scoped to the deal in view, nothing kept
- Interest
- An AI drafted expression of interest, editable
- Control
- Read-only: the AI never executes an investment
The starting point
A regulated marketplace where investors browse private investment deals, submit expressions of interest, sign documents and follow their holdings like a bank statement, and issuers raise rounds. It already ran as a .NET application with a front end built by the client's own team.
What had to change
Investors had to read long, inconsistent deal documents themselves, and issuers found potential investors by hand. The client wanted AI to help both sides act faster without ever deciding for them.
What could not break
- The AI is read-only: it never executes an investment or changes a compliance rule
- Answers come only from the deal in view and a few named sources
- Portfolio insights stay simple and never become financial advice
- Issuers accept or reject every expression of interest themselves
- The live application and its database are reached only through APIs
The calls we made
We framed every investor-facing agent as an advisor with a narrow source and no power to act.
- 01
Read-only by design
The investor agent studies profile and activity data to recommend deals and give simple portfolio insights, such as equity against debt, but it cannot place an investment or touch a compliance rule.
Ruled outLetting the agent execute investments or modify compliance rules.
- 02
Check the issuer's summary
When an investor opens a deal, the issuer agent's summary is shown with a second summary built on the investor side that checks it against current market data and the deal's positioning.
- 03
One deal at a time
The assistant answers from the documents uploaded for the deal in view and a few named sources, with in-memory retrieval scoped to that deal and nothing kept between sessions.
- 04
Drafts that people finish
The expression of interest is drafted by AI inside a five-step wizard and stays editable, and the issuer accepts or rejects it.
- 05
Scores, and the issuer decides
Investor matching scores candidates from the client's own investor profile file on investment type, ticket size, market cap and geography, and the issuer chooses whom to approach.
How the system fits together
What we built
Deal recommendations
An investor agent that recommends deals from each investor's interests, previous activity and risk profile, with simple portfolio insights such as equity against debt.
Investor-side market check
When an investor opens a deal, a second AI summary checks the issuer's summary, adds current market analysis and looks at the deal's competitiveness and positioning.
Deal-scoped assistant
A stateless chatbot over each deal's documents and a few named sources, using in-memory retrieval scoped to that deal.
Drafted expression of interest
A generator that drafts the expression of interest for the selected deal inside the platform's five-step wizard, editable before the investor submits it.
Investor matching for issuers
Finds and scores potential investors for an issuer from the client's investor profiles on investment type, ticket size, market cap and geography.
How it works
An investor signs in
Their deals page leads with recommendations drawn from their interests, previous activity and risk profile.
They open a deal
The issuer's AI summary appears with a second summary that checks it against the current market.
They ask questions
The assistant answers only from that deal's documents and a few named sources, and keeps nothing afterward.
They express interest
A five-step wizard ends with an AI drafted expression of interest they can edit before submitting.
The issuer decides
The issuer accepts or rejects each expression of interest and sees scored potential investors for the round.
A closer look
{
"investor": "inv-3107",
"interests": ["medical devices", "diagnostics"],
"risk_profile": "moderate",
"recommended_deals": [
{"deal": "deal-1042", "type": "equity", "why": "Matches the medical devices interest"},
{"deal": "deal-0987", "type": "hybrid", "why": "Matches the diagnostics interest"}
],
"portfolio_insight": "Holdings lean toward equity, with no debt deals yet.",
"mode": "read_only"
}What changed
The investor agents were delivered as backend APIs beside the live product. Every decision to invest, accept interest or approach an investor stays with people.
Under the hood
- Datics role
- AI added to a live product
- Data scope
- Retrieval is limited to the documents of the deal in view and a few named knowledge sources, with no memory across sessions or deals.
- Decision boundary
- The agents are read-only: recommendations, summaries, drafts and scores only, with no power to execute an investment or change a compliance rule.
- Integrations
- Backend APIs called by the client's front end, the client's investor profile data, Azure OpenAI and SerpAPI web search for market context.
- Deployment
- AI services on EC2 in a separate AWS environment beside the live application and its SQL Server database.
AI and models
Cloud and infrastructure
Apps and APIs
Also in the stack
How it shipped
Discovery and scope
Discovery calls and a scope of work set the rules: APIs only, a separate cloud environment, and recommendations instead of decisions.
Investor flow written out
We set out the investor journey step by step, from recommendations to a five-step expression of interest, and asked the client which fields AI should fill and from which source.
Investor agent demo
Investor matching from the client's investor profile file and an investor relations view for issuers were shown working to the client.
Chat on every agent
The client asked that each agent talk with users directly and ask for what is missing, so the investor chat followed the issuer chat.
Handover and acceptance testing
The agents went to the client's team as APIs for integration and acceptance testing.
Flows agreed before building
We wrote each investor flow out step by step and asked the client to confirm which fields fill themselves, and from where, before building.
Built in step with the issuer side
The investor agents followed the baseline set by the issuer's deal creation agent, and the investor chat was wired in after the issuer chat, so both sides of the marketplace work the same way.
What this unlocked
The client asked for ongoing support, estimates for the remaining agents in its plan and a move of the AI services into its own AWS account.
Related work
Have a product with the same kind of constraints?
Tell us what already exists and what you want it to do. We will tell you what it takes.