DavidAgents

DavidAgents Cortex

Tier two · Intelligence layer

Every event in your company. Governed, queryable, learning.

The data lake, analytics, and ML plane under everything: change-data-capture from every system into a medallion lakehouse, real-time data products out of the other end, and the same spine feeding your dashboards, your agents' memory, and your models.

The problem

“What happened last week?” should not be a quarter-long project.

Your data is in twelve systems, your analytics vendor has a copy of some of it, your ML team has a stale export of the rest, and nobody trusts any of the numbers. The fix is architectural, not heroic: capture every change at the source, land it once in governed layers, and let every consumer — dashboard, agent, model — read from the same truth.

Capabilities

Change-data-capture

Every system streams. Nothing is exported, ever again.

CDC pipelines watch every system of record in the box (and the ones you bring) and stream row-level changes into the lake as they happen. No nightly batch, no CSV archaeology — freshness is measured in seconds and pinned by tests.

● lake freshness has a live wall row

The medallion lakehouse

Bronze to silver to gold — with soft deletes and history

Raw changes land in bronze, conform in silver, and serve as gold current-state products with full history behind them. Deletions tombstone instead of vanish, schema drift is contained at the boundary, and every layer is queryable SQL.

● the org chart on our site reads gold

Real-time products

Dashboards and feeds that are seconds old, not days

Fleet health, revenue events, tool-call telemetry, customer timelines — materialized continuously and served over simple APIs. The proof numbers on this very page are one of them.

● this page's stats stream from it

The learning loop

Agent memory, evaluations, and ML on the same spine

Cortex is what makes the agents smarter: their memory distills from it, their evaluations score against it, knowledge search retrieves from it, and your models train on it. One data plane, compounding returns.

● agent recall + evals run on it daily

What's inside

The manifest.

  1. 01 CDC pipelines row-level capture from every source
  2. 02 Medallion lakehouse bronze / silver / gold, full history
  3. 03 Real-time OLAP seconds-fresh analytical queries
  4. 04 Gold data products materialized, API-served
  5. 05 Knowledge & vector search documents + embeddings, governed
  6. 06 Agent memory store the fleet's long-term brain
  7. 07 Evaluation loops agents scored on real outcomes
  8. 08 Tool-call telemetry every agent action, analyzable
  9. 09 Graph store entities + relationships for retrieval
  10. 10 ML training surfaces clean, versioned, yours

How it lands

Connect the sources

CDC connectors attach to your systems of record — box-native ones are pre-wired; external ones join through the same pipeline pattern.

Conform the layers

We shape silver and define your first gold products: the timelines, rollups, and feeds your business actually asks for.

Point the consumers

Dashboards, agents, and models read the same governed plane. New questions become queries, not projects.

Proof

On duty for us, right now.

The live numbers on this page, the fleet dashboards in our command center, and the memory our agents recall from are all served by the same Cortex instance, continuously CDC-fed from our production systems.

checks passing on this line failing last full run

Put Cortex on your payroll.

Same platform we run our own company on. First line is a purchase — the second is a decision.

See my data plane