Fetch company context and summarize likely workflows without inventing facts.
INTERNAL REFERENCE IMPLEMENTATIONRevenue Engine.
Revenue Engine.
Built before we sold it.
This is an internal VFSYSTEM control-plane implementation for B2B research, qualification, outreach preparation and reply handling. It is shown here as engineering proof — not as a client case study and not as evidence of customer results.
What is actually implemented
The system uses a persistent task queue and state store. Each stage writes state and events, and outbound actions require explicit approval rather than relying on model confidence.
Score fit for a custom AI operations engagement and keep uncertainty explicit.
Generate a concise draft using only the supplied research. No fabricated case studies or metrics.
A message cannot be dispatched unless its state is explicitly changed to approved.
Replies are polled, categorized and used to cancel queued follow-ups when appropriate.
Production rules encoded in the system
The dispatcher rejects any message whose state is not explicitly approved.
Outbound volume is bounded by configuration rather than left to an autonomous agent.
Follow-ups are created as drafts and must pass through approval again.
A reply stops the blind follow-up loop so a human can take over the conversation.
Research and draft prompts explicitly prohibit fabricated facts, case studies and performance claims.
Stale running tasks are detected and returned to the queue instead of being silently lost.
Implementation status
These labels describe the implementation itself. They are not customer performance claims.
SQLite queue and statePersistent prospects, tasks, messages, opportunities and delivery state.WORKING
Research → qualificationTested against the provider boundary.TESTED
Qualification → draftTested end-to-end in an isolated fixture.TESTED
Approval → dispatcherTested while remaining in dry-run.TESTED
Reply loopImplemented for contacted prospects.IMPLEMENTED
Agency control planeOpportunities, clients, projects, work items, artifacts and payments.IMPLEMENTED
The control boundary is code, not a promise
A simplified excerpt of the dispatch rule used by the internal system:
if message.status != "approved":
raise RuntimeError("message must be explicitly approved before send")
if sent_today >= MAX_DAILY_SEND:
raise RuntimeError("daily send cap reached")
result = send_email(message.email, message.subject, message.body)Bring one workflow.
We will tear it down.
Start with a real process, the systems involved and the outcome you want. The first conversation is about whether the problem is worth automating — not about selling you a generic AI platform.