All work
172017TRG Tech

Real-Time Social Sentiment Engine

Near-real-time brand sentiment from 100K+ social posts a day, at under 30 seconds end-to-end.

Before

Marketing customers needed near-real-time brand sentiment from 100K+ daily social posts across Twitter, Facebook, and Instagram — far beyond manual monitoring.

After

A streaming pipeline classifies sentiment and topic, detects spikes, and pushes alerts to subscriber dashboards. It processed 100K+ posts/day per tenant at under 30 seconds median end-to-end latency and onboarded 12 brand customers in the first quarter.

Challenge

Marketing customers needed near-real-time brand sentiment from 100K+ daily social posts across Twitter, Facebook, Instagram.

Approach

Streaming pipeline classifies sentiment + topic, detects spikes, and pushes alerts to subscriber dashboards.

How it was built

  1. 1

    Streaming ingestion

    Phase 1

    Built a queue-backed streaming pipeline (RabbitMQ) to absorb 100K+ posts/day per tenant across three social APIs without backpressure.

  2. 2

    Sentiment + spike detection

    Phase 2

    Classified sentiment and topic in-stream and added spike detection, because the alert on a sudden shift is what customers actually pay for.

  3. 3

    Multi-tenant dashboards

    Phase 3

    Pushed alerts into per-tenant dashboards, onboarding 12 brand customers in the first quarter.

Key architecture decisions

Streaming over batch

Why · 'Near-real-time' was the product. A queue-backed streaming design is what held median latency under 30 seconds at 100K+ posts/day.

Spike detection as the headline feature

Why · Raw sentiment scores are noise; the alert on an unusual shift is the signal customers act on.

Impact

  • Processed 100K+ posts/day per tenant
  • Median end-to-end latency under 30 seconds
  • Onboarded 12 brand customers in the first quarter
100K+/day
volume
30s
latency
12
tenants

What I'd tell someone building this

  • 01 · If real-time is the promise, architect for streaming from day one — retrofitting is painful.
  • 02 · Detect and alert on change; a static score is rarely the thing customers need.
  • 03 · Design for multi-tenancy early when the product is sold per-customer.

Tech stack

Node.jsRabbitMQSentiment AnalysisAngular

Ask anything about Real-Time Social Sentiment Engine

AI scoped to this project · GPT-OSS 120B