PortMem
PortMem · Pre-Seed · July 2026
Regulated firms are shipping AI-generated analysis they cannot verify or evidence. PortMem is the independent layer that checks every claim, names every omission, resolves which document version governs — and signs the proof.
Andre Bacellar, Founder · Live on GCP · 4 published papers · 01 / 12
KPMG citations checked out; report withdrawn
dedicated GenAI section, FINRA's 2026 Oversight Report
Compliance blocks deployment. The bottleneck: defensibility, not capability.
Four questions: Supported? What's left out? Which version controls? Can we prove we checked?
Generators — grade their own homework.
Checkers — blind to what's missing; an opinion, not evidence.
Archives — store, don't verify.
Governance — police the rules, not the work.
Independent verification with court-grade evidence. That is PortMem.
1Claim-level verdicts — supported / contradicted / uncertain / not found, with evidence quotes.
2Completeness — names each omitted material provision.
3Signed evidence — hash-chained, Ed25519-signed, verifiable offline.
4Which version governs — authority and supersession; the capability no vendor we examined publicly documents.
Upload the governing documents; versions and amendments handled.
Submit the AI-written report, from any tool.
Get the audit — per-claim verdicts, evidence quotes, omissions, human sign-off.
Export the proof — a signed evidence packet an examiner can verify offline.
Live on GCP production. Deployment tiers: cloud · single-tenant · air-gap.
| Benchmark | Claude Haiku, no documents | Same model + PortMem |
|---|---|---|
| FDA recall lookups · pharma · n=100 | 29% | 99% |
| SCOTUS overruling pairs · legal · n=100 | 48% | 99% |
| FASB ASC supersessions · finance · n=20 | 95% | 100% |
End-to-end QA accuracy, same model — the layer is the only variable, and PortMem returns the citation and controlling version; the baseline returns neither. Every number is artifact-backed at portmem.com/methodology, including the correction we published to our own earlier baselines — that discipline is the product.
| Claim verdicts | Completeness | Signed evidence | Version governance | |
|---|---|---|---|---|
| Norm Ai $120M C @ $1.2B · Jul 2026 · “the verification layer” | binary flag | checklists | WORM storage | — |
| Generators Harvey · CoCounsel · Hebbia | self-citation | — | — | — |
| Scorers Vectara · Patronus · Check Grounding | score only | — | — | — |
| Archives Smarsh · Global Relay · Purview | — | — | activity logs | — |
| PortMem | ✓ shipped | ✓ shipped | ✓ shipped | ✓ shipped |
Public vendor materials as of July 2026 · dated teardown in the data room
EU AI Act Art. 50 duties + GPAI fines apply
first dedicated GenAI section — our feature list, as an exam expectation
into the category in 90 days: Norm $120M · Patronus $50M · Hadrius $27M
Generators shipped first. Regulators published the expectations. The evidence layer is the gap — one auditor covers every tool.
Buyer — the CCO / GC blocking AI rollouts; a risk budget, not a productivity budget.
Wedge — a paid 30-minute proof-of-value on the firm's own documents.
Land / expand — metered verification; our surface grows with their AI adoption.
Channels — direct (17-target named pipeline) + OEM into the vendors who own the buyer.
FINRA member firms
SEC-registered RIAs
initial SAM, bottoms-up, knobs shown
published arXiv papers
security findings fixed same day
false contradictions, 3-domain run
named targets, warmth-ranked
Live on GCP — solo, bootstrapped.
Pilot search — no revenue yet; the round exists to fix exactly this.
90-day goals — 3 paid proof-of-values, 1 signed design partner, billing live.
published arXiv papers
years enterprise sales — Huawei, Cisco, Google
Founder — ARClogic · Columbia MBA · BSc EE.
Hire 1 · Q4 2026 — founding engineer: owns the evidence/ledger surface and deployments.
Hire 2 · Q1 2027 — compliance-native GTM: runs the pilot motion the founder opens.
paid proof-of-values · month 3
ARR · month 12
ARR path visible · month 18
Use of funds — ~45% hires · ~25% founder + ops · ~15% security/compliance · ~15% infrastructure + pilots.
What de-risks this check — the product is live, the evidence is auditable; the remaining risk is commercial execution, which the funds buy directly.
Talk to us.