SigSent — The Human Signal Lab

SigSent is an early-stage B2B SaaS product for testing messages before they reach the intended audience. Teams can test competing versions with real reviewers and compare what people did with what they said afterward.

I designed and built SigSent independently. Access is currently request- and invite-based. SigSent sits before the sending process; it does not replace an email service provider or post-send campaign analytics.

A test contains two to five variants — subject, preview text, body, and call to action. Each variant receives a tokenized review link. Reviewers open the link in the browser and read normally; no account, panel membership, or installation is required.

During the session, SigSent captures behavioral telemetry and then collects structured feedback after the review. Minimum-evidence gates control when a comparison becomes actionable: an early leader can be identified after three completed responses per variant, while five are required before the release gate can return anything beyond WAIT. The resulting release status is GO, CONDITIONAL, WAIT, or STOP, with the supporting evidence shown alongside it.

Behavioral telemetry includes opens, scroll depth at 25 / 50 / 75 / 100, dwell time, CTA interaction, completion, and abandonment — each associated with a specific reviewer session and variant.

Structured feedback captures trust, clarity, and action intent on a five-point scale, plus recall and an optional written comment. One response is accepted per session to prevent double-counting.

Gemini supports drafting, refinement, and evidence analysis, but model output remains separate from the release decision. Comparison thresholds are deterministic: the same qualifying evidence produces the same release status. AI-assisted recommendations are grounded in reviewer sessions, scores, and comments rather than model preference alone.

The client is built with React and TypeScript against a Node.js/Express API, with application data stored in MongoDB Atlas. Firebase Authentication handles account access through email and supported federated sign-in providers, while product access remains invite-gated.

Google Gemini supports analysis and assisted drafting. An optional MediaPipe Face Landmarker workflow runs locally in the account holder’s browser; camera data is processed on-device and is never uploaded. Reviewer sessions do not use camera or microphone access.

The frontend is deployed on Vercel and the API on Render. Reviewer access uses tokenized links rather than reviewer accounts, with expiration configured at creation for up to 30 days.

Reviewer sessions do not require a name or email, and the SigSent application does not store reviewer IP addresses or derive reviewer location. Hosting providers may process IP addresses as part of standard request handling and security. SigSent does not use advertising or cross-site tracking cookies.

Core service providers include Firebase, MongoDB Atlas, Google Gemini, Vercel, and Render. Consent-based product analytics use Google Analytics, and Stripe is used if and when paid subscriptions are enabled.