Retrieva — YC one-pager
AI-Native Compliance Infrastructure for financial entities. Category (YC Fall 2026 RFS): AI-Native Compliance Infrastructure.
One-liner
Retrieva is the AI compliance officer for DORA third-party ICT risk — it reads a financial entity's vendor contracts, security evidence and subprocessor chains, and continuously produces the risk assessments, concentration analysis and Register of Information the regulator requires. What a compliance team does across weeks of spreadsheets, Retrieva does continuously — with a human owning every decision and a full audit trail behind it.
The problem
DORA is mandatory, hard, and manual. Since 17 January 2025, every EU financial entity (banks, insurers, investment firms, fintechs) must run continuous ICT third-party risk management under the Digital Operational Resilience Act: assess every ICT provider, map nth-party subprocessor chains, identify concentration risk (Art. 28–29), maintain a machine-readable Register of Information (EBA RT.02.01), and keep exit strategies for critical providers.
Today this is stitched together with spreadsheets, siloed GRC tools, expensive consultants and a Chief Compliance Officer's headcount — and the cost grows faster than the business. It is exactly the AI-native problem the YC RFS describes: monitoring regulatory change, flagging anomalies, generating reports, keeping audit trails, across a jurisdictional patchwork.
The regulator proved the pain. In the 2024 ESA "dry run," each Register of Information faced 116 data-quality checks — only 6.5% of firms passed all of them. The top failures: incomplete contract data, missing subcontractor (nth-party) information, and incorrect Critical/Important Function classification — because "a spreadsheet cannot maintain the relational integrity between entities, contracts and functions." Meanwhile ~half of UK financial firms spent >€1M on DORA, and smaller firms lack the resources to map their supply chain at all. 93.5% of firms failed — that is the market. (Full analysis + sources: Product Vision → Market problem.)
The product
Retrieva ingests the documents a firm already has (contracts, SOC 2 / ISO reports, DPAs, subprocessor lists — including scanned and chart-heavy PDFs via multimodal ingestion) and:
- Assesses each vendor against the DORA control set (RAG over a curated DORA knowledge base).
- Maps the graph — critical/important functions → providers → nth-party subprocessors — and surfaces concentration risk (everyone on the same cloud) and single points of failure.
- Generates the Register of Information (RT.02.01) and audit-ready reports.
- Keeps a human in the loop — every AI-extracted edge or risk decision is confidence-scored and ratified by a person before it affects a compliance artifact.
Governed autonomy, not a chatbot: every model call is logged (audit trail), access is policy-gated, and the human owns accountability.
Why now
- DORA is in force (Jan 2025) and enforcement is ramping — demand is regulator-mandated, not discretionary.
- The EU AI Act adds AI-governance obligations on top — the same buyers, the same year.
- Intelligence got good enough — multimodal document understanding + RAG can now read the evidence base a firm actually has, not a clean API.
Go-to-market: France first → EU → world (incl. USA)
Phase 1 — France (beachhead). French financial entities are supervised for DORA by the ACPR (and AMF). We start here because the demand is mandatory, the language and regulatory context are ours, and the buyer (DSI / RSSI / conformité) is reachable. Land mid-size insurers, banks and fintechs that must comply but can't justify a large GRC build.
Phase 2 — European Union. DORA is a single EU regulation — the same product serves all 27 member states with only language + local-authority nuances (e.g. BaFin in Germany). One regulatory engine, a continent of buyers.
Phase 3 — World, including the USA. The core primitive — a continuously-updated third-party / nth-party risk graph tied to a regulator's register format — generalizes. We map the same engine to adjacent regimes: US TPRM (OCC / FFIEC third-party risk guidance, NYDFS 500), UK operational resilience (PRA/FCA SS1/21), and the broader global vendor-risk market. Start where the law forces adoption (France/EU), then export the engine to jurisdictions that need the same capability.
Why us — the moat
- DORA-native, not GRC-generic. The concentration + nth-party graph tied to the RT.02.01 register is the regulator's own hard question — competitors selling questionnaires can't easily copy it.
- Governed autonomy by construction. Retrieva runs on a full AI operating platform (LiteLLM gateway with EU residency, Langfuse audit trail, policy-gated access, human-in-the-loop). In regulated finance, auditable + compliant by construction is the only sellable form of AI.
- Already built and live. Retrieva runs in dev + prod today (multi-tenant, enterprise security, multimodal ingestion, the concentration graph) — a working product, not a deck.
The wedge → the vision
Retrieva starts as compliance infrastructure, but ICT third-party risk is the entry point to a larger arc: the governance layer of the AI-native insurer — a machine-first, human-governed operating model where AI runs the operational company and humans own the decisions. Compliance is the wedge with mandatory demand; the operating model is the expansion. (See the AI-First Insurance Operating Model vision.)
Status & the ask
- Live at retrieva.online (dev + prod), multi-tenant, DORA knowledge base, concentration graph, Register of Information, multimodal ingestion, full LLMOps observability.
- Serves double duty as the author's RNCP39583 certification project — battle-tested, documented.
- The ask: YC — to turn a working, regulation-mandated product into the compliance infrastructure every EU (then global) financial entity runs on.
Sources (market validation): EBA — 2024 DORA Dry Run key findings (6.5% / 116 checks) · Neotas — DORA TPRM 2026 · Infosecurity Magazine — compliance headwinds · Fintech Global — true cost of non-compliance · Kiteworks — DORA TPRM guide · Trusty — shared compliance burden