Reliable, Smart Workflows for Finance, Wealth and Insurance

8 Просмотры

Спасибо! Поделитесь с друзьями!

Вам не понравилось видео. Спасибо за то что поделились своим мнением!

Издатель Oct 17, 2026

“Fill out a form for Mr. Harper” becomes a process that lasts days: identify the right household, find missing information, contact the client, wait, follow up, obtain a signature and retain the evidence. Karthic Rao uses that wealth management example to show why a fixed flowchart breaks on ambiguity and an improvising agent creates a different reliability problem. At Perpetual, his design puts control flow, evidence and external actions in a deterministic system. Models handle interpretation and judgment at explicit, typed points. A judgment can create a smaller plan, but that plan must pass the same compilation and approval boundaries.

The planner translates a request into a semantic graph whose nodes point back to the request. A compiler validates it against the firm's action contracts, returning a valid plan, a clarification request or a refusal. Approved plans are immutable, and external actions have separate gates tied to the exact payload. Waiting and followups are state machine steps; uncertain action outcomes go to manual review rather than automatic retries. Rao shows an illustrative 20 decision run with 19 decisions assigned to a smaller model and one to a frontier model. That ratio is a design target, not a measured fleet result. Typed, logged judgments offer a path toward privately hosted inference and future training, while the training loop remains forthcoming.

Speaker info:
- https://github.com/hackintoshrao
- https://www.linkedin.com/in/hackintoshrao/

Timestamps:
00:00 - Reliability and intelligence in regulated workflows
2:19 - One request becomes a multiday process
5:44 - Ambiguity, approvals and evidence
6:45 - Drawing the deterministic boundary
8:01 - Typed planning and a compiler that can refuse
9:56 - Payload approvals, waits and child plans
11:45 - State machines and uncertain side effects
12:52 - Smaller model targets and the private inference roadmap