At Titanium Birch, investment decisions draw on structured processes—reviewing memos, evaluating opportunities, modelling frameworks, and building a shared understanding of strategy over time. DealDuck is the internal application that supports those workflows, turning messy real-world inputs into structured material the investment team can reason about clearly.
What DealDuck does
DealDuck manages processes related to direct venture capital investments and fund investments. It is a document-processing and workflow engine inside Trunk, TB’s primary monorepo, that takes in business documents arriving in inconsistent formats and turns them into structured, comparable data through agent-driven extraction and classification.
The application covers several distinct workflows: reviewing investment committee memos, triaging new deal opportunities, conducting strategy interviews, and maintaining the investment strategy itself. Each workflow is encoded as a structured process—a defined set of instructions that specifies how the work should be done, what the inputs and outputs look like, and where human judgement is required.
TB uses DealDuck for both fund investments and direct venture capital (VC) investments, but the main workflows are not the same for each. The fund-investment path centres on investment committee (IC) memo readiness: getting a fund commitment memo into shape before it goes to the IC, using the structured review process below. The VC path centres on deal triage for new opportunities (intake through to a pass-or-continue decision) and on portfolio framework modelling—the versioned strategy document, sleeve definitions, and related analysis that keep VC portfolio construction explicit, as described later in this article.
Investment committee memo reviews (fund investments)
Before a fund investment goes to the investment committee (IC), someone has to assess whether the memo is ready. DealDuck includes a structured review process specifically for this: evaluating an investment memo for IC readiness and producing actionable feedback.
The process is not a rubber stamp. It evaluates whether the memo states the thesis upfront, covers the key risks, presents internally consistent analysis, and provides sufficient supporting evidence. Reviews are saved as structured artefacts, which means the feedback is traceable and comparable across different memos and different time periods.
This matters because memo quality directly affects decision quality. A memo that omits a key risk or buries the thesis in background material makes the committee’s job harder. By encoding what “IC-ready” looks like as a repeatable, structured process, DealDuck helps maintain a consistent standard without relying on any single person to enforce it.
How this shows up: IC readiness work often involves surfacing gaps in a memo—what is missing, what is unclear, what does not line up. DealDuck runs the memo against that bar and records the result as structured feedback. The investment team can then spend its time tightening thesis and evidence against those flags instead of chasing informal comments across different channels.
Deal triage (VC investments)
When a new startup investment opportunity arrives, it needs to be assessed quickly: is this worth spending time on, and if so, what do we need to learn next? A recurring decision the triage workflow supports is whether an opportunity moves from intake to partner review (or is parked with a clear reason).
DealDuck’s deal triage workflow takes an opportunity from intake through to a pass-or-continue decision. It works against a due diligence checklist that defines what information is needed and what the evaluation criteria are. The workflow is designed to surface the most important questions early rather than leaving them to emerge late in a lengthy review.
The triage process handles the fact that incoming materials are rarely clean. A deal might arrive as a PDF memo, an emailed appendix, and a slide deck with key numbers buried in images. DealDuck uses markitdown to convert PDFs into markdown, making the content machine-readable without losing the structure. The application treats all external material as untrusted input and isolates processing before anything enters internal workflows.
The engineering tension here is familiar to anyone who has built extraction pipelines: the input formats are inconsistent enough that no single parser handles everything, but the output needs to be consistent enough for structured comparison. Getting that right—and making it obvious where the system is uncertain rather than quietly inventing confidence—is an ongoing challenge.
Illustrative triage record (schematic, not from a real deal): the workflow is aimed at producing a small, comparable set of fields the team can scan before deeper work. Placeholder labels only—no real companies, funds, or figures.
| Field | Example shape (illustrative) |
|---|---|
| Opportunity ID | OPP-0000 (internal reference) |
| Thesis fit | Short free-text flag, e.g. aligned / weak / unclear |
| Stage | Label such as seed / A / B / other |
| Geography / sector tags | Controlled vocabulary or short notes |
| Material gaps | Bullet list of missing items from the checklist |
| Open questions | What the team would need to answer next |
| Suggested next action | e.g. request follow-up materials, partner review, pass with reason |
VC portfolio framework modelling (VC investments)
Beyond individual deals, DealDuck supports the broader question of how TB’s venture capital portfolio should be constructed. The application maintains a strategy document that contains the current investment policy, sleeve definitions, and return targets.
Framework modelling is about making portfolio-level decisions explicit and testable. Rather than carrying strategy around as institutional memory—where it drifts as people join and leave—it is written down, version-controlled, and linked to the analysis that informed it.
The strategy document is supported by interview transcripts and their interpretations, each stored as a separate file and linked from the main document. When the team conducts strategy interviews to refine their approach, the process follows a structured guide that shapes the conversation and captures the output in a consistent format. Over time, this builds a documented trail of how and why the strategy evolved.
How it fits together
DealDuck’s workflows share a common pattern: take unstructured or semi-structured input, apply a defined process, and produce output that is structured enough to support good decisions. The technology does not replace human judgement—it structures the preparation so the humans can focus on the actual call.
The application is still being developed and refined. Some workflows are more mature than others; the IC memo review process, for instance, has a well-defined skill and a clear output format, while other areas are still being shaped. We track the broader engineering tensions behind that unevenness in The open questions. That is expected for a system this early in its development. The value is not in having a finished product but in having a consistent, improvable process that gets better as the team learns what works.