Consultant (NDA) · Capital markets · Agentic AI
Agentic AI built for enterprise deployment
I designed and built the autonomous AI agents that triage exchange and vendor change and incident notices for capital-markets infrastructure, then built the SSO, integrations, and audit trails that let a regulated buyer's IT and procurement teams sign off on the rollout, not block it.
Role
Software engineering consultant
Domain
Capital markets
Engagement
Independent · under NDA
Outcome
Deployed in a regulated enterprise environment
(T-DIAGRAM, abstracted schematic, no real UI, no client data)
The business problem
Hundreds of notices a day, triaged by hand
Exchanges and data vendors publish a constant flood of change, incident, and maintenance notices — spec updates, connectivity changes, policy circulars, outage alerts. Any one of them might break a firm's trading or market-data operations, or might be noise. Teams read the inboxes, judged the impact, and opened tickets by hand: slow, error-prone, and impossible to scale to the volume.
The answer was autonomous agents — and I built them: agents that ingest each notice, extract the structured facts, judge the real-world impact, and route it to action. But the wall was never the model. It was everything around it — identity, integrations, and whether a regulated buyer could trust an automated decision enough to deploy it. So I built that layer too.
How I approached it
Make the autonomy adoptable, not just impressive
Building agents that triage notices accurately was the core of the work — but a working demo is the easy half. I also built the three things that decide whether enterprise teams approve a rollout: how the platform authenticates against their identity, how its integrations behave on the third retry, and whether every autonomous decision an agent makes can be explained and audited after the fact.
In capital markets, an automated action without an audit trail is a liability, not a feature.
What I built
The agents — and the layer that made them deployable
I built the autonomous AI agents at the core of this platform, then built the SSO, idempotent integrations, and audit trails around them. The agents are the product; the fundamentals are what let a regulated buyer actually deploy them.
1 · The AI agents
I designed and built the autonomous agents that do the work: ingest heterogeneous exchange and vendor notices, extract structured operational data, classify real-world impact against a firm's own trading and market-data footprint, and route the result into the ticketing the team already uses. Grounded in a structured service taxonomy and each customer's configured logic, so a decision reflects that firm's actual operations — and built explainable and auditable, so every step a regulated team can interrogate.
2 · Enterprise SSO
Enterprise SSO over SAML 2.0 and OpenID Connect / OAuth 2.0, with per-tenant IdP onboarding — metadata exchange, certificate handling, role and attribute mapping from each customer's identity provider — and multi-tenant isolation from the first commit. In this market, SSO isn't a feature request; it's what enterprise IT teams verify before a rollout happens.
3 · Resilient integrations
Idempotent integrations to downstream ticketing and the inbound sources upstream — built to survive retries, partial failures, rate limits, and inconsistent upstream data. Onboard a new provider or destination without re-architecting the core. The value wasn't the happy path; it was staying trustworthy when the third party didn't.
I built the AI agents and owned the SSO and integration layers as part of the platform team. Not the whole product, and I'll say so.
The outcome
Adoption, not a demo
Deployed
The SSO, integrations, and audit trail made it deployable in a regulated enterprise environment. The agents run in production.
Manual → autonomous
Inbox-driven monitoring replaced with autonomous triage of the daily notice volume, surfacing only what affects the firm.
Auditable
Every autonomous decision traceable — status, revision flags, and a full event timeline, the way a regulated environment requires.
No numbers here are mine to publish. What I can say is the bar it had to clear, and that it did.
Built with
- Autonomous AI agents
- Agentic workflows
- SAML 2.0 / OIDC
- OAuth 2.0
- Idempotent integrations
- Multi-tenant
- Audit / event timeline
This is what an agentic-AI engagement looks like.
Have an AI problem that has to work in production?