AI-powered pre-call sales intelligence — and a LinkedIn network analyzer — both running on Cloudflare's edge.
Signal-Stack is a B2B sales-intelligence platform built around a single proposition: the rep should walk into every meeting with a brief that actually reflects the relationship — not a CRM summary, not a LinkedIn scrape, an honest synthesis of what's been said and signaled across every channel.
It does this by reading the last 90 days of calls, the last 60 days of email, and any signals available in Salesforce, HubSpot, LinkedIn, and Clay. The synthesis runs through Claude using an org-customizable prompt template, and the resulting brief lands as email or in Slack the morning of every meeting on the calendar.
A brief that reflects the relationship.
Not a CRM summary.
Calendar → Connectors → Signals → Claude → Brief
Exports → Overlaps → ICP match → Shortest warm path
Four connectors fan in (Gong, Gmail/Outlook, LinkedIn, CRM), get hydrated in parallel, and feed a signal-computation layer that scores deal health, relationship velocity, champion stability, and deal-velocity trend. Claude synthesizes those signals into a structured brief at the org's prompt template.
The entire stack runs on Cloudflare — Workers, Pages, D1, KV. No origin servers, no Docker, no Kubernetes. Cold start measured in single-digit ms.
Every connector credential (Gong API key, Salesforce OAuth token, etc.) is AES-GCM encrypted before storing in D1. Decryption is fail-closed.
Independent peer review by OpenAI, all P0/P1 findings closed, security headers + CSRF + rate limiting throughout.
Six-stage flow — contact classification → Gong hydration → Gmail analysis → parallel enrichment → signal computation → Claude synthesis. Each stage independently testable.
Gmail scope minimized to brief generation only. Restricted-contact list. Erasure log. CCPA + GDPR data-subject request flow.
Signal-Stack is where the theses on agentic workflows, edge-native AI products, signal computation, and revenue-team context get tested against real users and real workflow pressure.
The advisory work argues for orchestration, signal computation, and edge-native architecture. Signal-Stack proves those theses against paying customers, not slide decks.
Most AI in sales generates more emails. Signal-Stack does the opposite — it classifies, summarizes, scores, and routes inside the workflow the rep already lives in. The brief shows up where the meeting lives.
Building exposes the real trade-offs behind the strategy — connector quality, prompt brittleness, latency, error budgets, billing edge cases. Each one sharpens the advisory.
Signal-Stack is the operator counterpart to the advisory work — the place where the theses on agentic measurement, vector-based audience matching, and edge-native B2B software get tested against paying customers and real data.
It's also the home of the adcp.signal-stack.io subdomain — the production demo of the AdCP Signals Adaptor, a separate open-source reference implementation contributed back to the working group.
Ask about the essays, frameworks, playbooks, or how to work with Evgeny.
