Portfolio Intelligence

Portfolio intelligence that asks the questions first.

Threadbase reads everything your portfolio produces, links it into a knowledge graph with full provenance, and surfaces conflicts, dependency drift, and risks before anyone thinks to ask. The analysis is done for you.

The problem

The dangerous question is the one nobody knows to ask.

Portfolio failures usually trace back to information nobody connected — two teams running on contradictory assumptions, a dependency quietly growing while everyone treats it as settled, a risk split across three orgs so no one sees the whole of it.

Status tools only answer the questions you already know to ask. The signal exists — it's just scattered across decks, meetings, and a few long-tenured people. By the time someone thinks to ask, it's already a problem.

Portfolio
The full set of initiatives, projects, and bets an organization has running at once.
Analysis, done for you
Threadbase runs continuously and brings findings to you — there's no dashboard to interrogate.
How it works

Threadbase asks the questions first.

It reads everything the portfolio produces, compares what one team said against what another assumed, and surfaces what's wrong before anyone asks — continuously, with no dashboard to query.

1

Read everything

Agents ingest every document the portfolio produces — decks, PDFs, spreadsheets. Azure Content Understanding and a vision pass turn complex layouts into structured facts.

2

Link with provenance

Each fact becomes a claim with confidence, temporal windows, and full lineage — joined into a knowledge graph on Postgres and pgvector.

3

Surface before anyone asks

An agent reasons over the whole corpus — not scheduled detectors — and proactively flags conflicts, dependency drift, and expired assumptions.

The Q3 migration is blocked until the data-residency review clears, which pushes the EMEA launch into Q4.

Traces to Q3 Program Review · slide 14
Evidence first

Nothing enters the system without a receipt.

No claim enters Threadbase without a link to the exact document, page, and region it came from. Every finding traces back to its source — so a human verifies in one click, not one hour, and nothing the system says is taken on faith.

Built by

Winslow Tandler

Threadbase is a working case study in evidence-first AI systems — multimodal document understanding, a provenance-linked fact ledger, a knowledge graph on Postgres and pgvector, and an agent layer with durable memory reasoning over the whole corpus. In active development, built end to end. The full write-up — the problem, what shipped, and how each piece fits — lives on winslowtandler.com.

Next.js React 19 TypeScript Postgres + pgvector Azure Content Understanding Agentic knowledge graph
Stay in the loop

Follow the build.

Threadbase is being built in public. Drop your email for occasional updates — new capabilities, write-ups, and milestones. No noise.