Applied AI · systems used in active work
Three operating systems, not three prompts.
I have worked hands-on with generative AI since ChatGPT’s public
launch. The proof is not a subscription or one clever prompt—it
is the operating system around the model: role context, trusted
sources, repeatable instructions, quality checks, human
approval, and documentation.
I use these systems in real finance-media work, then package the
working process so another person can learn it, run it, and know
when to escalate.
System 01
Research → publishing
Congressional Trading Content OS
Built a documented pipeline for high-risk finance content: scan
CapitolTrades and Quiver, verify the official filing, add market
and committee context, score the story, draft it, run
fact-and-voice checks, then require human approval.
- Source hierarchy and verification rules
- Story scoring matrix and production queue
- Prompt library, reply router, and brand guardrails
-
Prohibited claims, escalation triggers, and final sign-off
System 02
Parallel research
Four-desk investigation workflow
Designed four role-specific workspaces—Source, Investigation,
Signal, and Publishing—to divide large questions, reconcile
conflicting claims, score relevance, and turn verified findings
into decision-ready work.
- Claim-and-source tables with confidence labels
- Parallel workstreams and contradiction checks
- Watchlists, scoring rubrics, and publish checklists
- One finding adapted into X, email, briefs, and reports
Output proof: six long-form finance and
political investigations totaling 241 pages.
System 03
Analysis → decisions
Operational analysis and multi-model QA
Use AI to structure long Telegram histories, email metrics,
social performance, customer feedback, and sprawling
research—then pressure-test the conclusion across ChatGPT,
Claude, and Gemini before reducing it to a scorecard, case
study, or next action.
- Trade-record and campaign-performance reviews
- Audience, offer, funnel, and monetization analysis
- Cross-model critique instead of single-answer trust
- Human review of sources, math, claims, and final output
AI workflow design & team training
A workflow is not finished until someone else can run it.
I turn a working system into a team playbook: setup, role
context, trusted sources, prompt sequence, example inputs and
outputs, prohibited claims, QA checklist, escalation rules, and
human sign-off. Training can include a live walkthrough, guided
practice, documented handoff, and an adoption scorecard.
- 01Map the role
- 02Build context
- 03Set source rules
- 04Test on live work
- 05Document
- 06Train
- 07Measure
Ask for a live workflow walkthrough
↗
Selected internal systems built from active Bors Finance and The
Fifth Signal operating needs. They are not autonomous agents: final
outputs require source verification and human approval. Workflow
documents can be reviewed during an interview. “Team training”
describes Ben’s ability to package and teach these workflows; it is
not a claim that he trained or fine-tuned an AI model.