Human-reviewed AI workflow

AI-SDR speeds up outreach without handing judgment to the model.

A campaign workflow for research, draft generation, review, regeneration, follow-up, and analytics. The human stays in charge of final actions.

React dashboard Cloud Functions Firestore Claude Browser E2E Prompt Director Lead Finder Sender safety

RoleFull-stack builder across backend, dashboard, and browser E2E checks.

WorkflowCampaigns, prospects, generated drafts, approvals, follow-ups, and analytics.

StatusWorking system, deployed path, and active dogfooding.

BoundaryNo unsupervised final sends or account-sensitive actions.

The product call

Most outreach automation optimizes for volume. I wanted an operating loop.

The useful shape is not "AI sends messages for me." It is campaign context, prospect research, draft generation, review, regeneration, logging, and follow-up in one place.

That framing matters because outreach is a trust-heavy workflow. Automation should remove admin friction, not turn a social account into a risky black box.

AI-SDR offerings screen with reusable service offer cards linked to campaigns — Offerings: reusable service positioning, customer criteria, safe claims, and campaign links.

AI-SDR LinkedIn profiles screen with sender limits, business hours, and warm-up controls — Sender safety: connected LinkedIn profile, daily limits, business hours, timezone, and warm-up controls.

AI-SDR Lead Finder builder with AI-generated ICP filters — Lead Finder builder: generated ICP filters from the offer context, with accepted changes before launch.

AI-SDR Prompt Director test conversation interface — Prompt Director: a testing surface for refining outreach prompts before campaign execution.

AI-SDR saved Lead Finder screen with reusable prospect searches — Saved Lead Finders: reusable prospect searches attached to campaigns, with import and run history.

AI-SDR prospects table with LinkedIn status and ICP scores — Prospects: imported leads, LinkedIn status, ICP score, pending state, and next action flow.

AI-SDR prompt library with reusable outreach prompt variants — Prompt library: reusable outreach prompt variants that can be edited, tested, and attached to campaigns.

What I built

A full-stack workflow around approval and state.

Campaign setup and context capture so generated drafts know the offer, audience, and constraints.
React dashboard surfaces for campaign state, prospects, conversations, analytics, and review queues.
Prompt Director for improving outreach instructions, testing model behavior, and keeping generation quality inspectable.
Claude-assisted message drafting with edit, approve, reject, and regenerate loops.
Browser E2E checks around approval/regeneration, while final LinkedIn sends stay outside automation.

Trust boundaryThe system helps draft and organize. It does not secretly publish or send.

Workflow designEvery AI action sits inside a queue, state transition, or review moment.

Product relevanceUseful where AI has to support judgment, not replace the person doing the work.

Takeaway

The interesting engineering is the boundary design.

AI-SDR is useful because it treats AI as a drafting and decision-support layer. The product value is in the queue, the state, the recovery paths, and the human approval gates.

Back to portfolio