Diligence decision-support experiment
A decision-support prototype for venture diligence
A full-stack diligence-workflow prototype with deal, research, scoring, portfolio, and reporting surfaces. Its source demonstrates workflow architecture and provider abstraction; it does not establish investment performance.
Business ↔ technology bridge · Diligence decision-support experiment
Explores a reviewable full-stack diligence workflow with visible assumptions and replaceable model providers.
This is a portfolio and source interpretation—not a customer, deployment, certification, or outcome claim.
Deal, market, founder, technical, portfolio, and reporting evidence is assembled across disconnected data sources and individual judgement.
Separate workflow, APIs, persistence, retrieval, scoring, and model providers, including a deterministic mock path for repeatable testing.
Deal, research, diligence, scoring, portfolio, and report surfaces across FastAPI, Next.js, data, retrieval, and provider-adapter boundaries.
Missing sources, conflicting signals, score assumptions, and reviewer decisions return to thesis, diligence scope, and portfolio monitoring.
Source evidence
A full-stack diligence-workflow prototype with deal, research, scoring, portfolio, and reporting surfaces. Its source demonstrates workflow architecture and provider abstraction; it does not establish investment performance.
Repository evidence is not client, deployment, certification, or outcome evidence.
Architecture
Three source boundaries: workflow UI, modular APIs, and a replaceable model-provider layer with deterministic test support.
Source map only: a layer can represent implemented code, an adapter boundary, or documented design. It does not assert a configured production service.
The Next.js source represents pipeline boards, founder review, scoring, competitive mapping, and portfolio-monitoring surfaces. Source presence does not establish live data coverage or investor adoption.
The FastAPI source represents modules for deal sourcing, due diligence, founder scoring, technical assessment, competitive intelligence, term-sheet drafting, and post-investment monitoring. Stage-calibrated logic — what matters at seed is not what matters at Series C.
Provider adapters represent Anthropic, Mistral, Ollama, Hugging Face, and deterministic mock paths. Model behaviour, locality, and data handling depend on the selected provider and deployment configuration.
The Problem
Why VC due diligence is broken
Deal evidence is often assembled manually across founder, market, technical, commercial, and regulatory sources, making provenance and review consistency difficult to preserve.
Quality varies by analyst. No standardised framework means critical signals get missed on some deals.
Weeks to produce a single investment memo. Hot deals close before diligence is complete.
Insights from past deals aren't reused. Each evaluation starts from scratch with no institutional memory — and EU-specific data (EIC grants, BPI France, INPI filings, AMF disclosures) is completely ignored.
Features
Source surfaces for a model-assisted due diligence workflow
Source scope only: an entry can describe implemented code, a documented adapter, or a configured workflow. It is not a statement that a production service is active.
01
The source represents a founder-assessment rubric covering background, track record, network, technical depth, leadership, resilience, market insight, and integrity — weighted 25% of the Kairos Score. The investor remains responsible for the conclusion.
02
Workflow surfaces for TAM/SAM/SOM assumptions, timing analysis, and market-opportunity scoring. Source availability depends on configured data adapters.
03
The source represents an AI-drafted competitive landscape analysis with positioning, differentiation assessment, and moat evaluation. The investor remains responsible for the conclusion.
04
Adapter and workflow surfaces for digital presence, repository activity, hiring signals, customer sentiment, filings, and commercial datasets. A named source is not evidence of live access.
05
The source represents fundraising-timing, change-signal, key-person-risk, and valuation-assumption workflows. Their inputs and outputs require licensed data, calibration, and investor review; no predictive accuracy or investment outcome is claimed.
06
The source assembles draft investment memos, diligence reports, competitive views, and risk registers for investor review. No turnaround or decision-quality outcome is claimed.
8 scoring dimensions
The prototype makes scoring dimensions and rubrics explicit so an investor can inspect assumptions instead of treating a model output as a verdict.
Repository surface only: availability, completeness, performance, and operational fitness require separate validation.
DIM 1Domain depth × execution velocity × team-fit signals (co-founder dynamics, prior shipping evidence, hiring track record). Inputs represented in source: LinkedIn enrichment, public-commit cadence, and interview transcripts. A named source is not evidence of live access.
DIM 2Bottom-up TAM with sources, top-down sanity check, 5-year growth thesis, regulatory tailwind/headwind. The proposed rubric would penalise 'all of the EU is our market' with a structured why-now requirement.
DIM 3Captures user-research artefacts, retention evidence where available, and the gap between a demonstration and an operable product. The investor remains responsible for the conclusion.
DIM 4Revenue curve quality (not just absolute number), net dollar retention, sales-cycle compression over time, organic vs paid mix. Stage-calibrated — what matters at seed is not Series A.
DIM 5Burn multiple vs stage, payback period on CAC, runway clarity, prior round price-to-quality ratio. The proposed rubric would flag 'too much capital for the milestone delivered.'
DIM 6Technical assessment by domain, drafted by the LLM backend with different rubrics for Seed and Series C. It is an input for the investor to inspect, not a verdict on the founder's claimed novelty.
DIM 7Data moat depth, switching cost analysis, regulatory moat where applicable (e.g. clinical, defence), network-effect curve shape. The 'why does this not get crushed by Big Tech?' answer.
DIM 8Sales motion clarity, ICP discipline, channel quality, founder-led-sales transition plan. The proposed rubric would penalise 'we'll hire a VP Sales to figure it out' answers.
Tech Stack
Named technologies identify repository dependencies or adapter surfaces; they do not assert a live integration or deployment.
Source Code
The Kairos VC prototype source code is available upon request for evaluation and partnership purposes.
To access the source code, please sign the Non-Disclosure Agreement. This is a standard legal formality.
Looking for AI-powered due diligence or startup scaling expertise? Explore my consulting services.
Partnership
For teams exploring a reviewable, model-assisted diligence workflow.
From demo to dependable operation. Subscribe for new Physical AI reading.
Unsubscribe anytime. No spam, ever.