BetterBrain×ViridonHow it fits together

You're not trading cost for customization.

Everything we deliver for Viridon is fully bespoke — built for your documents, your process, your use cases. We're faster and cheaper because we bring a foundation across the whole stack, not because anything is off-the-shelf. This walks through how the pieces fit: the knowledge layer, the tools, the orchestrator, the workflows, and the mini-apps your teams actually touch.

The team stops at 99. The capability doesn't.

01 — The mental model

It isn't a layered cake. It's a network of modules, built in parallel.

The instinct is to picture this as sequential layers — buy the knowledge layer, then bolt on custom tools, then a UI on top. That model breaks, because you can't know what the knowledge layer needs until you know the workflow it has to serve. So we build top-down and in parallel: many small, reusable modules, wired together as the use cases demand.

The trap Sequential layers

UI  — custom interface on top
Custom tools & workflows
Off-the-shelf knowledge layer

Implies "one vendor sells the bottom layer cheap, we add the specific bits on top." It assumes the foundation is generic and fixed. In practice you hit a wall — the retrieval layer alone never satisfies a real workflow, and you can't spec it until you know the workflow.

How it actually works Modules in a mesh

KL·A
KL·B
KL·C
KL·D
Tool
Tool
Orch.
Your data

Each layer is a set of plug-in modules, not a slab. A workflow reaches across layers and pulls only the modules it needs. We build the modules once, then recombine them — which is exactly why a new use case is fast instead of a fresh ground-up build.

Foundation — already built by BetterBrain, customized for Viridon Built bespoke for Viridon Viridon-owned data & sources

02 — The architecture

The whole picture, end to end.

Here's how we think about the architecture. Your documents feed a Knowledge Layer made of many components. Tools and an orchestrator draw on those components — many-to-many. The orchestrator chains tools into workflows. Workflows compose into the mini-apps your teams use. Color shows what we bring vs. build net-new — but everything on this diagram is fully customizable. Every tool, every knowledge-layer component, workflows, orchestrator, mini-apps — even the pieces we've built before, we customize and tune to exactly what Viridon needs.

Flow ↑ Sources (bottom)  →  Knowledge Layer  →  Tools  →  Workflows  →  Mini-apps (top)

Click any tool, workflow, or mini-app to pin what it uses across the stack below. Click again to clear. The links are many-to-many — that's the point.

1

Viridon Sources

your data, owned by you
SOURCESharePoint · proposals · RFIs · contracts · ISO/RTO public docsStays in your tenant. Ingested into the knowledge layer; never used to train other models.

feeds
▲   ingested & structured from   ▲
2

Knowledge Layer

far more than indexing & retrieval — many components
all appsKL·AIndexing & RetrievalHybrid BM25 + vector (Vespa). This is all Glean gives you.
all appsKL·BIngestion & StructuringAny doc type → parsed, chunked, tagged, structured records
KL·CSection & Question ExtractionBreak proposals into sections / RFP questions
KL·D · bespoke"What Changes" MapWhich fields, numbers & vendors shift each cyclefor you
KL·ESelection-Report AdviceWin/loss themes mined from ~200-pg sponsor reports
KL·FTemplate GenerationAuto-built proposal templates from past wins
all appsKL·GConcept Mapping & LinkagesConcepts, customers, projects, terms — and how they link
all appsKL·HScoped Retrieval + RerankingProject / client / global scope, RBAC-aware, reranked
KL·I · bespokePublic-Doc EnrichmentISO/RTO transmission plans, deliverability studiesfor you
KL·J · bespokeRFI Q&A + SME DelegationPrior RFI answers & who owned what last timefor you
KL·K · bespokeStandard-Terms PlaybookViridon's clause library & acceptable positionsfor you
KL·L · bespokeOnboarding GlossaryCompany context, tutorials & how concepts connectfor you

powers
▲   tools & orchestrator pull components   ▲
3

Orchestrator & Tools

both read the knowledge layer · marked tools serve every app
all appsT·QGrounded Q&A / chatCited answers over the knowledge layer — the MCP entry point for every team
T1Read & comment on a paragraphSuggests improvements vs. selection-report themes
T2Draft a sectionFrom template + structured prior wins
T3Identify opportunitiesWhere to differentiate this bid
T4Flow updates across 300+ pagesNumbers, vendors, names everywhere
T5Evaluate against criteriaScore a draft vs. what wins
T6Aggregate & match attachmentsSME reports into one narrative voice
all appsT7Web research & scrapeLive external + public-doc context
T8Build a templateAuto-derive from past proposals
T9SME routerLikely owner from past delegation patterns
T10RFI trackerAuto-populate items & assign owners
T11Clause & field extractorCounterparty, dates, term, obligations
T12Screen vs. standard termsFlag only what needs human review
T13Contract trackerRepository of every NDA & agreement

chains
▲   orchestrator chains tools   ▲
4

Workflows

orchestrated or deterministic chains of tools
WF · proposalSetupTemplate + flag what's outdated
WF · proposalStrategyFind angles & framing
WF · proposalDraftingAI teammate drafts & pulls in attachments
WF · proposalEvaluationReview, comment, propose edits
WF · RFIIntake & matchParse questions · find prior Q&A
WF · RFIDraft responsesFrom proposal, SME reports & past RFIs
WF · RFIRoute & trackAssign SME owners · populate tracker
WF · ISO/RTOSME Q&APublic docs + project history for a customer
WF · onboardingNew-joiner chatCompany context, people, terminology

composes
▲   composed of   ▲
5

Mini-apps

what each team touches
APP·1Proposal Writing AssistantOrigination · Erin
APP·2RFI Response DrafterOrigination
APP·3Legal Contract ScreenerLegal (2-person)
APP·4ISO / RTO SME + OnboardingAll teams

Two things the colors say. First, the components marked all apps — retrieval, ingestion, concept linkages, scoped retrieval, grounded Q&A, web research — are shared infrastructure every mini-app reuses. Second, almost everything is foundation we've built and customize to you; the bespoke pieces — app-specific tools (T1, T3, T4, T6, T10, T12) and knowledge-layer modules KL·D & KL·I–L — are where we spend the saved time. Glean would bring only KL·A, some generic off-the-shelf tools (not customizable), and stop.

03 — The grammar

A workflow is just a recipe of these parts.

Once you see the modules, every workflow is a combination of them. Some need the orchestrator to reason and route; some are a fixed, deterministic chain of tools. Either way, it's parts off the same shelf.

Workflow  =  Orchestrator (optional)  +  Tool·A Tool·B  +  KL·A KL·C

Worked example — "Selection-report suggestions" workflow

Orchestrator + T1 · read & comment T3 · identify opportunities + KL·A retrieval KL·E selection advice KL·G concept mapping

The orchestrator reads the selection-report advice, pulls the relevant prior sections, runs the comment tool paragraph-by-paragraph, and surfaces where this bid can differentiate. Swap the parts and you get a different workflow — no new ground-up build.

04 — The first mini-app, end to end

Proposal Writing Assistant: four phases, one foundation.

Erin's real workflow runs over months. Here's how the modules above show up across it — Setup is mostly her, then the AI works as a teammate that drafts, researches, comments, and proposes changes she approves.

Phase 1

Setup

Erin-led · deterministic
  • Erin highlights outdated info in a past proposal that must change
  • Inserts new project basics — name, sponsor, key numbers
  • AI generates the working template & flags everything likely to change
T8 templateT4 flow updates KL·CKL·DKL·F
1
Phase 2

Strategy

AI + Erin · orchestrated
  • Work through each section against the client brief & relevant docs
  • Use selection feedback + project-specific docs to find angles
  • Surface areas of opportunity from knowledge of Viridon's process
T3 opportunitiesT5 evaluateT·Q Q&A KL·AKL·EKL·GKL·H
2
Phase 3

Drafting

AI teammate · orchestrated
  • Multiplayer document editor with a writing assistant
  • Aggregates SME attachments into one narrative voice
  • AI drafts, comments, researches & scrapes the web — like a teammate
T2 draftT6 aggregateT7 researchT1 comment KL·AKL·BKL·CKL·FKL·H
3
Phase 4

Evaluation

AI + human gate · orchestrated
  • Evaluate the full proposal; leave comments across it
  • Human approves / denies each task or change the AI proposes
  • AI takes on approved tasks and flows the edits through
T5 evaluateT1 commentT4 flow updates KL·DKL·EKL·GKL·H
4

05 — The same foundation, more mini-apps

Every future use case reuses what we've already built.

The whole point of the foundation is leverage. Proposal writing is its own beast — but the next mini-apps don't inherit its proposal-specific tools. They sit on the shared infrastructure every app uses, then add a few pieces of their own. That's why each new app is a fraction of the first.

Shared foundation · used by every mini-app KL·A retrieval KL·B ingestion KL·G concept mapping KL·H scoped retrieval T·Q grounded Q&A T7 web research

RFI Response Drafter

Origination teams

Drafts answers to customer follow-ups using the proposal, SME reports, and prior RFI Q&A; flags the likely SME for open items from past delegation patterns; auto-populates and assigns owners in an RFI tracker.

Shared foundation
KL·AKL·BKL·GKL·HT·Q
Specific to this app
KL·J RFI Q&A + delegationT9 SME routerT10 RFI tracker

Legal Contract Repository & Screener

2-person legal team

Pulls counterparty, dates, term, and obligations from every NDA/contract into a tracker; screens incoming NDAs against Viridon's standard terms and flags only what needs human review.

Shared foundation
KL·AKL·BKL·HT·Q
Specific to this app
KL·K standard-terms playbookT11 clause extractorT12 screen vs. termsT13 contract tracker

ISO / RTO SME Assistants

Development & origination

AI SMEs with deep knowledge of specific customers — answering questions on project histories, long-standing system constraints, and past workshop outputs from large volumes of public ISO/RTO documents.

Shared foundation
KL·AKL·BKL·GKL·HT·QT7
Specific to this app
KL·I public-doc enrichment

AI Onboarding Assistant

All new joiners

Internal chatbot covering company context, projects, people, industry terminology, and how core concepts connect. Almost entirely shared foundation — which makes it simple to stand up once the core components of the brain exist.

Shared foundation
KL·AKL·BKL·GKL·HT·Q
Specific to this app
KL·L onboarding glossary

06 — The leadership case

Off-the-shelf isn't a cheaper version of this. It's a different product.

Two decisions for leadership: buy off-the-shelf or build the platform — and if we build, with whom. Below is the outcome each path actually delivers, measured against everything the platform needs. Same skeleton in all three; what's filled in is the whole story.

Pre-built & optimized by BetterBrain Built bespoke Generic off-the-shelf Not delivered — and no one in-house to build it
vs off-the-shelf (Glean)

You'd buy roughly a tenth of the solution — the search box — and still have no way to build the other nine-tenths.

vs other vendors

The identical bespoke outcome — but pre-built, optimized layers make it faster, more performant, and far cheaper.

07 — The same picture, component by component

Same bespoke result. A fraction of the build.

Three ways to get here. An off-the-shelf tool gives you one box and stops. A bottom-up vendor builds every box from scratch — fully custom, but slow, expensive, and untested. We bring the boxes already built and spend our time slotting in the few that are uniquely yours.

$

Glean / off-the-shelf

SaaS · fixed
ABCD EFGHI
  • Indexing & retrieval over your data — and that's it
  • No "what changes," no selection-report advice, no templates
  • Can't be shaped to Erin's real workflow
  • You still can't build the rest yourselves
$

BetterBrain

foundation + fully bespoke
ABCD EFGHI
  • Bring most of the stack already built — and proven
  • Spend the saved time on the pieces unique to Viridon
  • As bespoke as a ground-up build, fast enough to pilot in weeks
  • Fully owned by Viridon — sits in your platform, transfers with the company
$$$

Bottom-up build

new vendor · from scratch
ABCD EFGHI
  • Fully custom — but every box built from zero
  • Months of build before anything works end to end
  • Untested architecture; implementation scars come on your dime
  • ~$300K-class effort for the same destination
Brings off the shelf (built & proven, customized to you) Builds bespoke for Viridon Not available

You get the $$$ result at the $ price — and you own everything that makes Viridon, Viridon.

We're not cheaper because we cut corners on customization. We're cheaper because the foundation — UI, tools, orchestrator, knowledge layer — is already built. We're not writing all the code; we're slotting in the pieces that make this Viridon's. The knowledge layer, apps, and orchestration sit in your platform and transfer with the company — an owned asset in the data room.

Foundation across the full stack. Delivery fully bespoke.

08 — Engagement & team

BetterBrain

Breadth across the stack. The right people, when it matters.

You get strategy, engineering, quality, UX, commercial, and governance in one engagement — with steady leads on the ground and specialists we bring in as each part of the project needs them. Senior throughout.

Day-to-day delivery· engagement, build & quality
Abhishek Bhargava
Abhishek Bhargava
Engagement lead · AI strategy
CMU computer science & computational finance; ex-YC, commodities trading. Roadmap and priorities — your main point of contact.
Darshan Vanol
Darshan Vanol
Forward-deployed eng · full-stack
Builds the knowledge layer, retrieval and app backends in your environment.
Ilona Litvinova
Ilona Litvinova
Quality & evaluation eng
Integraton and regression testing, evals, UX improvements.
Relevant Specialists· matched to the work
DK
Dima Kyrychukflex
Forward-deployed eng · app & UX
Mini-app interfaces and the AI editing workspace.
Alex Brogan
Alex Broganflex
Commercial & exit-strategy
Ex-Goldman IB — finance-related project initiatives (i.e. cost containment tooling).
Michael Boyer
Michael Boyerflex
CIO advisor · IT governance
Enterprise IT, security and compliance.
Retrieval & knowledge layer
Darshan + Abhishek
Security & compliance
Abhishek + Michael
BetterBrain · Prepared for Viridon One engagement, many disciplines — we bring in depth wherever the project needs it.
09 — Depth behind the bench
BetterBrain

Meet the experts. Built to deploy, not just advise.

Research-grade AI depth meets operators who've shipped in production — across finance, robotics, enterprise IT and energy. We're backed by leading funds, and by individual investors and advisors from the very companies building the models and data platforms Viridon will run on.

Academia & research

Carnegie Mellon University
Stanford University
UC Berkeley
MITRE

Enterprise & finance

Goldman Sachs
PwC
CCI
Y Combinator

Industry & operations

Perdue
ESC · Environmental Systems Co.
CapSen Robotics
Backed by
Funds
SAMSUNGNEXT
hustle fund
Individual angels & advisors — people who work at
OpenAI Snowflake

Individuals investing & advising in a personal capacity — not corporate investments by these companies.

Core capabilities
StrategyRoadmap & prioritization · use-case discovery · eval design · governance & risk.
ImplementationKnowledge layer & retrieval · workflow orchestration · context-aware agents · proposal & document automation.
Industry expertiseFinance & project finance · legal & compliance · manufacturing & logistics · energy & infrastructure.
BetterBrain · Prepared for Viridon Academic depth and production scars — the same team that builds the foundation ships it in your environment.