One platform for research, intake & analysis

From eight silos to one platform

Eight apps had grown up separately, each building its own version of the same machinery. We consolidated around five moves:

• Built the engine once — Ford. Research, document breakdown, verification, matching, and cost control now live in one shared engine. Each app invokes it with its own "persona pack" (prompts, tools, output shape, budget). TaskFlow and LP Flow already run on it in production.
• Gave knowledge one home — Magellan. Every research paper, external deck, and filing lands on one shelf, read deeply once, and compounds — instead of being re-read in whichever app it happened to land.
• Gave people & firms one spine — Webster. One registry links every person and firm across the CRMs: 776 people and ~892 firms that appeared in two books are now one linked entity each. Both CRMs show "also in the other book" badges, and enrichment is never paid for twice.
• Made news flow, not pile — Dow. One news collector; the daily stream ages out, and the durable facts (deals, IPOs, sector stats) publish automatically onto Magellan's shelf.
• Kept one voice — Franklin. The writing room reads the shelf and the news flow through a live bridge — no second copy of anything — and keeps only what's ours: themes, drafts, evidence, voice.

The same machinery had been built three to five times

A full audit of the portfolio — 60-plus features across seven codebases — found the same capabilities written again and again, with no shared code between them. Each row is one capability and the number of separate implementations of it.

The apps — each with one job

Under the hood — shared by every app

Apps on top, one engine underneath, shared spines below

Every fact on the platform carries its source and as-of date. Findings are staged, human-approved, and undoable — the model proposes, people commit.

How a PDF is broken down

Drop in a research report, an S-1, a benchmark book, or an external deck — three doors in, one pipeline, and the knowledge lands everywhere it's relevant.

Magellan's report page — the way every breakdown reads across the platform.

The charts worth keeping come out as exhibits

A 120-page report holds maybe eight charts you'd actually put in front of an LP. The engine picks them, says why each one matters, and cites the page — so they're ready to drop into a deck instead of buried in a PDF.

Eight exhibits lifted from the Hamilton Lane 2025 Market Overview — AI picks, each removable; a removed pick never comes back.

Everything we've read, in one place, with its credibility on the label

Every report lands on one shelf and is read once — filed by type and credibility tier, with its analysis, facts, and exhibits attached. This is what "the firm's research memory" actually looks like.

The shelf: tier-1 reports, sector notes, internal work and press, each one click from its full breakdown.

Ford: the research engine, built once Live

Ask any app to research a person, a firm, or a question — the same engine answers, wearing that app's persona.

• One engine, many personas. LP Flow researches an LP with its evidence rubrics; DealFlow enriches a company; TaskFlow runs a background brief — same core, different persona pack. The persona changes the questions, never the quality bar.
• Two ways to invoke it: a button in the app for interactive jobs, or a schedule for background passes (overnight research, list backfills, monitoring).
• Everything is staged. The engine returns findings with confidence and sources; a person reviews and commits. Human edits always win.
• Cost is governed, not hoped about. Budgets per job, tiered model use, and a spend ledger — every AI dollar is attributed to the surface that spent it.
• The email door. Forward anything to ford@rivent.dev — the engine classifies it and routes it to the right place on the platform.

A live engine pass: staged findings, sources attached, one click to commit.

Webster: one spine of every person and firm Live

The same investor appears in the LP book, the deal book, a research report, and a podcast episode. Webster makes those one entity — linked, never merged.

• Link, never merge: one registry row per real person or firm, linked to each app's record with a confidence score. Links are undoable; the registry never writes into any app.
• Visible payoff: "Also in the other book" badges in both CRMs — the deal team sees when a VC is also an LP relationship, and vice versa.
• Never pay twice: before spending on enrichment or research, every app checks the registry's freshness first.
• Every extraction lands on the right entity: a quote about a firm in a report, a deal in a newsletter, and the CRM record all point at the same spine entry.

The registry surfacing inside the CRM.

Suggested matches wait for a human yes.

Dow: the news flows through and the facts stay Live

One collector reads the newsletters, deal announcements, and market news every day. The stream is for reading; the durable facts compound.

• One news system of record. Bespoke per-source readers turn each newsletter into structured sections with the verbatim passage behind every item — including charts lifted from emails as exhibits.
• Durable facts publish to the shelf automatically. A deal announcement becomes a signal on that company's Magellan dossier; a stat worth keeping becomes a datapoint with its source; IPO information always publishes, including full S-1 breakdowns.
• Scoped by relevance. Private, SLW-relevant companies and sector trends compound; mega-cap noise stays in the feed and ages out. Anything we already track always gets through.
• Franklin reads the same stream — its writing topics are fed by Dow's news, filtered by topic, so nothing is collected twice.

The daily read: structured sections, source passages one click away.

Franklin: writing with our voice

When we write — letters, decks, posts — Franklin pulls the evidence from the whole platform: quotes from the shelf, news from the flow, exhibits from the reports we've read.

• The bridge. Franklin reads Magellan's shelf live — searches its pages, quotes, datapoints, and key slides in place. No second copy of the research library anywhere.
• Evidence sweeps. Building an argument collects verbatim quotes into supporting, challenging, and context lanes — with the counterpoints made explicit, not buried.
• Exhibits on tap. The key slides Magellan pulled from external research appear as ready-to-use exhibits while writing a theme or deck section.
• The writing brief. "I want to write about X" → a what-we-have / what's-missing panel: quotes, datapoints, and exhibits per leg of the argument, with one-click engine research passes to fill the gaps.

The writing room: our argument, with the receipts attached.

Slide Library: every slide we've ever made, findable

The other half of writing is not starting from scratch. Every deck we've produced is broken into its slides, titled and categorized by AI, and searchable by what's actually on the slide — so the right exhibit is a search away instead of a memory exercise.

• 10,000+ slides across 120 decks, each one an object you can find, reuse, and drop straight back into PowerPoint.
• Categorized the way we think about a deck — investment thesis, track record, portfolio construction, market overview, case study — with AI-written titles describing what each slide argues.
• Honest about age: slides carry a stale flag when their underlying numbers have moved on, and near-duplicates are marked rather than silently multiplying.
• Distinct from Magellan's key slides on purpose: these are ours, those are the market's — and eventually one search will cover both.

Investment-thesis slides across the library, each titled by what it argues.

Podcast1: what was said, turned into facts

Hours of podcasts a week become a page you can read in two minutes: the claim each episode makes, the evidence behind it, the numbers, and every company named — extracted, not summarized away.

• One thesis per episode, then the claims behind it, so an hour of audio is scannable and quotable.
• Companies and people are extracted and resolved against Webster — so what a podcast said about a firm lands on that firm's dossier, not in a transcript nobody re-reads.
• Grouped by theme — regulation, AI, macro, talent — so a topic's coverage across every show reads as one body of evidence.
• The one app that keeps its own processing by choice, but publishes its output into the shared knowledge layer like everything else.

103 analyzed episodes; the open brief shows the claim, the twelve points behind it, and the twelve companies named.

New capabilities now land as personas, not new apps

Because the engine, the shapes, and the registry are shared, each new capability is a persona pack and a surface — not another silo.