Intelligent Actuaries

An AI lab for actuarial science.

We build agentic systems where actuarial science meets frontier AI — and deploy them, under private mandate, inside the perimeter of tier-1 insurers, reinsurers, sovereign and pension funds. Nine specialist agents simulate, predict, advise, and explain across reserving, mortality, capital, climate, pricing, pensions, and regulatory. Every output is traceable from data to submission pack, with FSCA, IFRS 17, and SAM citations built in.

This Hugging Face organisation is the public face of the Intelligent Actuaries Research Lab — preprints, reproducible simulators, and foundation-model baselines live here. The institutional workbench itself is private, by introduction.

---

The roster

Nine specialist agents, four capabilities each — simulate · predict · advise · explain — orchestrated by a router that hands work to the right specialist and stitches the answer back together.

---

What we've shipped

• 🌍 951m USD CLIMADA AAL anchor — a reproducible AAL on a global tropical-cyclone exposure that locks down the climate & cat pipeline.
• ✅ 429 tests in CI — mypy --strict, ruff, biome, and Bun tests gate every merge to main.
• 🧠 9 / 9 specialist agents live — one SSE wire format across the workbench; router and specialists ship together.
• 📜 Regulatory-aware by default — FSCA · IFRS 17 · SAM. Functions whose output lands in submission packs are pinned with @regulatory; commits to them require an explicit sign-off marker.
• 🔬 Foundation-model baselines — TimesFM-style benchmarks against classical actuarial methods. Domain models (Lee-Carter, Mack 1993, Vasicek/CIR) remain the defensible primary methodology; foundation models are a sanity check, never a substitute.

---

Stack

Polyglot by design. Python is the glue; heavy compute drops to a Rust crate via PyO3, R via rpy2, and Julia via juliacall. Hermes Agent is the primary LLM provider — Ollama and Claude swap in with a single config flip.

---

Engagement

Intelligent Actuaries is delivered through private institutional mandates, deployed inside the client perimeter, under a Master Engagement Agreement. We accept a small number of new counterparties each year — chosen so that every engagement is one we can defend in front of your statutory actuary and your prudential supervisor. Fees are quoted privately, per mandate, after introduction.

There are no public binaries, no marketplace listing, no community edition. Research artefacts on this Hugging Face organisation (preprints, reproducible simulators, foundation-model baselines) are published openly; the production workbench is engaged separately.

• Engagement · mandate · NDA · fee: [email protected]
• Press · introductions · partnerships: [email protected]
• Research lab · preprints · simulators: [email protected]

---

Where to find us

---

Built in South Africa 🇿🇦, for the actuarial questions that move balance sheets.