Tal Agmon
עבריתTell me about it

Engage / NOD137

Engage prepares the reply. A person approves it.

Engage is an independent R&D project for what happens after a social post is published: collecting comments, understanding intent, preparing reply options and routing potential leads. Its central boundary is structural—no reply can be published without explicit human approval.

My roleProduct architecture · Full-stack · Safety design
StatusBeta · Live staging
TrafficProduction customers not enabled
Real Engage beta documentation portal showing the project status and human-approval boundary

01 · The operating flow

From a public comment to a deliberate human decision.

Useful questions and potential leads can disappear inside comment streams. Engage is designed to reduce that operational load without allowing an automated system to speak publicly on a brand’s behalf.

  1. SyncRead configured LinkedIn and X engagement.
  2. ClassifyIdentify intent, confidence and sentiment.
  3. SuggestPrepare grounded reply options for review.
  4. ApproveA person accepts or edits the reply.
  5. PublishA worker handles only approved replies.
  6. RouteSend qualified leads to connected systems.
The boundary is code, not marketing language

The suggestion service, classifier, polling worker and MCP tools cannot publish by themselves. The publishing worker only receives replies that entered the approved state through an explicit approval action.

02 · What is implemented

A complete beta pipeline with the risky action kept behind a gate.

The implemented work currently includes:

  • LinkedIn and X connector implementations with normalized engagement records
  • Intent, confidence, sentiment and recommended-action classification
  • Two to three grounded reply suggestions with a deterministic safety screen
  • Explicit approval, durable publish jobs and retry handling
  • Qualified-lead routing through signed webhooks and Slack-format delivery
  • Brand-scoped API keys, idempotency and an append-only outcome trail
  • REST, OpenAPI and MCP interfaces using the same domain rules
  • A responsive React documentation portal for integration and review

Technical foundations

The service is built in TypeScript with Fastify, Zod, Prisma and PostgreSQL. Redis-backed BullMQ workers separate polling, classification, publishing and webhook delivery. The documentation portal uses React, Vite and Tailwind CSS.

Current beta status

The workflow is implemented and running in staging for integration validation. Production customer traffic is not enabled. Provider access, connector readiness and the production operating model are still being tested, so this page does not claim production adoption or customer results.

03 · Evidence

The visual story comes from the working project.

Engage beta project overview
The current public documentation identifies Engage as independent R&D in beta, with production customer traffic disabled.
Engage six-stage workflow with human approval highlighted
The implemented flow from sync and classification through approval, publishing and lead routing.
Engage animated end-to-end pipeline diagram
The developer portal exposes the same pipeline as integration documentation.
Engage human approval invariant and publishing boundary
The documented publishing boundary: the approved-reply row and publish worker are the only path to a public reply.
Visit the live Engage documentation

Start with the problem

A few lines are enough.

You do not need a finished brief. Tell me what is unclear, disconnected, slow, or ready to become something real.

Tell me about it on WhatsAppEmail contact@talagmon.com

You speak directly with me. I use your details only to reply to this enquiry.

WhatsApp Tal