Infrastructure

Pipeline Infrastructure

The threat engine runs on dedicated cloud workers behind a durable workflow engine, with a durable database for state, an in-memory cache for queues, and vector search for the Evidence Vault.

Overview

Signal verification and thesis monitoring need durable, replayable workflows — not stateless lambda calls. Every pipeline run, every drift check, and every stop-loss watch is a durable workflow with explicit state, timers, retries, and an audit trail. Workers run on dedicated cloud hardware.

Infrastructure features

Durably orchestrated workflows

Thesis Guard, Earnings Distillation, and stop-loss watchers each run as a durable workflow with deterministic replay.

Auto-scaling worker pools

Workers scale on signal volume. Earnings days and macro shocks burst capacity automatically; idle hours scale down to baseline.

Isolated tenant compute

Each user gets an isolated worker queue and isolated workflow namespace. Evidence is namespaced per user in the vector index.

Edge + cloud split

Scraping and browser automation runs on distributed edge nodes. LLM inference, embeddings, and DB writes run on dedicated cloud workers.

Stack reference

Each pipeline run uses the following components:

Workflow engineDurable workflow engine (per-user isolation)
WorkersDedicated cloud workers + edge nodes (scraping)
State storeDurable database (Diary, watchlists, threat records)
Queue / cacheIn-memory cache & queue
Vector storeVector search (Evidence Vault)
Event busWebSocket push for the terminal · MCP for agents

Operational targets

Resumable

Workflows survive crashes (durable workflows)

5 min

Scan cycle across sources

100%

Workflow audit trail