PFN Dai · Issue 01 · Memo to Capital
Capital, allocated with intent.
PFN Dai fuses transformer-based on-chain analytics with proprietary DeFi signal models so funds deploy capital 3.4× faster and surface risk 48 hours before the market reacts. Our clients don’t guess yield — they engineer it.
- Deep Flow v3.2
- 14 chains
- sub-180ms
- AUC 0.87
Operating scale, documented on a single baseline — not a marketing grid, an evidence ledger.
- 184institutional clients
- 27countries served
- 14EVM + non-EVM chains
- 4.2Mon-chain signals / sec
Chapter I · Methodology
Deep Flow — the research asset at the centre of the book.
The Deep Flow transformer is PFN Dai’s proprietary on-chain behaviour model. Trained on 2.1 billion labelled wallet trajectories spanning lending, perps, intent systems, bridges, and restaking, it scores the structural fragility of any address, contract, or pool in real time. It is not a generic LLM wrapper. It is a domain-specific architecture with custom positional encodings for transaction graphs and a forecasting head calibrated for sub-72-hour horizon risk events.
In independent backtesting, Deep Flow v3.2 achieves an AUC of 0.87 on rug-pull prediction at a 72-hour pre-event horizon — a setting where naive heuristics and public dashboards converge near chance. Signal latency from chain head to actionable alert sits below 180 milliseconds across all fourteen supported networks, sustained through dedicated validator-node infrastructure rather than third-party RPC.
Specimen metric — AUC 0.87 @ 72h · p99 latency 178ms · 2.1B+ labelled wallet behaviours.
Chapter II · Audited Performance
Performance, not promises — audited by Armanino LLP.
An 18-month independent examination of client accounts on PFN Dai against passive DeFi benchmarks. The conclusion: yield engineered with intention, verified by a third party.
- +6.8% average APY outperformance vs. passive DeFi benchmarks Armanino LLP · 18-month window
- 3.4× faster capital deployment from thesis to on-chain position vs. manual treasury workflows
- 48h pre-event risk signal horizon on Deep Flow v3.2 AUC 0.87 @ rug-pull prediction
- 180ms p99 signal latency across 14 EVM + non-EVM chains Dedicated validator-node infrastructure
Chapter III · Trust
Institutional-grade by every audit that matters.
-
Compliance posture
SOC 2 Type II certification achieved February 2023. ISO 27001 certification achieved June 2024. MiCA-aligned EU data residency — the only on-chain analytics provider combining both. PFN Dai is an analytics provider, not a broker-dealer.
-
Capital backing
Series A closed at $24M in March 2024, led by Polychain Capital with participation from Framework Ventures and Circle Ventures. Founders: Lin Dai (ex-Bridgewater quant) and Felix Pfander (former Google Brain researcher).
-
Client roster
184 institutional clients across 27 countries, including 9 of the top 50 crypto-native funds by AUM. Headquartered in Singapore; engineering office in Zurich. Team of 41 FTEs, 64/36 engineering-to-GTM.
-
Recognition
Named “Best DeFi Analytics Platform” at the HedgeWeek Digital Asset Awards 2024. Shortlisted in the Forbes Fintech 50 — Top 10 AI in Fintech, Q4 2024.
Chapter IV · Next Step
Open the platform. Walk through your own book.
A 45-minute working session with our research team. Bring a position, a pool, or an open risk question. We will route it through Deep Flow live, on your data, in your jurisdiction.
Or write directly — [email protected] · +65 6817 4421