A major player managing asset portfolios across Europe needed to modernize how they marketed deals to investors. Until now, every step of the investor communication process was manual.

We built an AI-powered platform that compresses this entire workflow from days to minutes. The team provides the raw asset data, and the system handles the rest: it extracts structured information using LLMs, generates branded investor teasers, composes natural-sounding emails, manages Q&A with built-in NDA awareness, and follows up with non-responding investors automatically.

The Challenge

Marketing assets large scale involves a surprisingly manual and repetitive set of tasks. The client's investment team faced several bottlenecks.

Assembling asset information is slow. Key data — lists, financial figures, asset specs — lives across Excel files, PDFs, and internal documents in the dataroom. Extracting the right numbers and formatting them into a presentable teaser can take hours per asset.

Investor communication doesn't scale. Each investor receives a personalized teaser email tailored to their profile — investment focus, preferred regions, language. Writing these individually is time-consuming, and template-based alternatives sound robotic.

NDA classification is high-stakes. Some information can be shared freely, while other details (investment agreements, investor names, certificates) require a signed NDA. Getting this wrong is a compliance risk, and investment managers carry this classification logic in their heads.

Follow-ups fall through the cracks. Managing dozens of investors and follow-up sequences across a team requires coordination that breaks down when done manually. Investors go cold simply because no one remembered to circle back.

The Solution

The team provides the data. The platform handles everything else — from extraction to investor follow-up — through five integrated modules:

1 — Intelligent Data Extraction. The system reads dataroom documents and uses LLMs to extract 18 structured asset fields: project name, asset specifications, annual return, costs, price, and much more. Every field includes a confidence score and source reference, so the team knows exactly where each number came from.

2 — Automated Teaser Generation. Extracted data flows into a branded template that produces an investor-ready teaser with asset details automatically selected from the dataroom — ready to attach to an email.

3 — AI-Composed Outreach Emails. The system uses an LLM to compose natural-sounding emails based on the asset data and investor profile. It adapts language, personalizes content, and generates compelling subject lines — producing emails that read like they were written by a person, not a machine. 

4 — NDA-Aware Q&A. When investors respond with questions, the system classifies each one in real-time: does this require NDA-protected information? If not, it answers from available documents. If it does, it requests NDA signature first. Full conversation history ensures context-aware responses. 

5 — Automated Investor Follow-Up. The system tracks engagement and automatically re-engages non-responding investors with personalized follow-up emails that adapt tone and content based on the investor's profile and prior interactions.

Architecture

The extraction pipeline handles document parsing and LLM-based data extraction with smart caching. The email module manages personalization and follow-up sequences. The Q&A module handles NDA classification, conversation history, and response generation. All three share a common LLM client, storage layer, and configuration system.

A Few Technical Highlights

Data Extraction with Confidence Scoring

The extraction pipeline converts documents into structured text and sends them to an LLM with a validated schema. Every extracted field returns a confidence score and a source reference, letting the team quickly verify low-confidence values rather than reviewing everything manually. Smart caching ensures re-extraction only happens when source documents change.

Dual-Layer NDA Classification

The primary layer is an LLM classifier that understands nuance. It also catches intent-based requests like "send me everything." A keyword-based fallback layer provides fast classification when the LLM is unavailable. The dataroom itself is organized into pre-NDA and post-NDA folders, giving the system a structural foundation for access control.

Natural Language Email Generation

The email composer builds a rich context from asset data and investor profiles — investment focus, preferred regions, language, company type — and generates emails that feel personally written rather than template-filled. Each email includes a compelling, non-spammy subject line tailored to the specific deal and investor.  

Tech Stack

AI & LLM: Cloud-hosted GPT models, AI-powered document processing, LangChain, LangGraph

Backend: Python, Pydantic, Pandas

Cloud & Integration: Object storage, email and calendar APIs, planned e-signature integration for NDA workflows

AI Applied to Reality

This project shows what becomes possible when AI is applied to a clearly defined business workflow rather than treated as a generic productivity layer. By embedding domain logic, compliance rules, and communication standards directly into the system, the client transformed investor communication from a manual, high-friction process into a fast, reliable, and scalable capability. The result was not just efficiency, but higher consistency, lower risk, and a markedly better experience for both investment teams and investors.

At Singularity, this is how we approach applied AI: working closely with client teams, grounding solutions in real operational constraints, and building systems that integrate seamlessly into existing ways of working. The focus is always on durable impact - solutions that continue to create value long after the initial build.

Call to Action

If your organization is facing similar bottlenecks in deal marketing, investor communication, or document-heavy workflows, we’d be glad to explore what targeted AI can unlock in your context.

Get in touch to discuss how Singularity can translate complex processes into scalable, high-impact AI systems.