VelaReach
MetaMinds — Enterprise Ad Platform
The Challenge
Today's media buyers juggle 5+ tools to manage ads: Keitaro (€49-499/mo) for tracking, Dolphin Cloud ($49-99/mo) for multi-account management, FabFunnel (~$99-299/mo) for campaign management, Triple Whale ($129-2K/mo) for attribution, and Northbeam ($1K-15K/mo) for Media Mix Modeling. Each has its own vocabulary, its own metrics, and its own weak spots. Buyers end up as human ETL pipelines, reconciling spreadsheets instead of making decisions. No existing platform ships all 10 ad platforms, full campaign CRUD, 11 attribution models, and AI creative generation in one place.
The Approach
Engineered VelaReach as a microservices-first enterprise platform designed to outperform every competitor on features at a fraction of their price ($99-299/mo). Built 9 independently-deployable services (8 Python + 1 Go) orchestrated through Kong API Gateway, each with health checks, Prometheus metrics, and Sentry error tracking. Paired them with a 35-page Next.js 15 + React 19 dashboard on Vercel using ECharts for visualization, Clerk for auth, and Zustand for state. Backed by ClickHouse OLAP for sub-second analytics on billions of rows.
System Architecture
Frontend on Vercel + 9 backend microservices on Hetzner + ClickHouse OLAP. Every service is independently deployable with its own health checks, Prometheus metrics, and Sentry error tracking.
Next.js 15 SPA Dashboard
35-page single-page dashboard built on Next.js 15.3 with React 19, TypeScript 5.7, and Tailwind CSS 4. Clerk authentication, TanStack Query 5 for data fetching, Zustand 5 for state, ECharts 5.6 for visualization, Radix UI primitives, React Hook Form + Zod validation. Deployed on Vercel with edge caching.
9 Microservices on Hetzner
FastAPI (Python 3.12) and Go (1.22) microservices orchestrated through Kong API Gateway. 8 Python services + 1 Go platform-connector, each independently deployable via Docker on Hetzner. Kafka for async event streaming between services.
ClickHouse OLAP Engine
ClickHouse columnar OLAP processes 111 million rows in 1.1 seconds. Materialized views for pre-aggregated KPIs. SSE streaming for live dashboard updates. Sub-second queries on billions of rows enabling real-time attribution, cohort analysis, and anomaly detection.
The Solution
- Built 9 independently-deployable microservices: analytics, auth, campaign, creative, budget, automation, ai-pipeline, notification, platform-connector
- Analytics service (port 8003): 24 routers, 50+ endpoints covering attribution, MMM, cohorts, fraud, anomaly, A/B testing, creative cockpit, reports, S2S tracking, P&L
- Auth service (port 8006): Clerk JWT verification, multi-tenant RBAC with 6 roles and 19 permissions, Stripe billing with 6 webhook handlers, audit logs
- Campaign service (port 8010): Full CRUD + bulk operations + CSV import on all 10 ad platforms
- Creative service (port 8002): Asset management with Shopify/WooCommerce product feed integrations
- Budget service (port 8005): AI cross-platform allocator + 15-minute pacing monitor with auto-pause on overspend
- Automation service (port 8004): Rules engine with 14 operators, IF/ELSE logic, background evaluator, safe-mode for manual approval
- AI pipeline (port 8008): FLUX 1.1 Pro image gen via Replicate, Claude ad copy in 30+ languages, video generation, performance prediction (CTR/ROAS/CPA), conversational chatbot
- Notification service (port 8009): Kafka consumer delivering to Email/Slack/Telegram/Webhook channels
- Platform connector (Go, port 8007): 10 ad platform integrations with OAuth2, encrypted token vault, per-platform rate limiting, unified models, Kafka publisher
Feature Breakdown
Attribution & MMM
- 11 attribution models: Markov Chain, Shapley Values, Bayesian, LSTM, Thompson Sampling
- Proprietary ML attribution ensemble (beats Triple Whale + Northbeam)
- Bayesian Media Mix Modeling with budget optimizer
- 365-day attribution windows with VTA configuration
- Path visualization for customer journey analysis
- Incrementality testing with synthetic control + power analysis
AI Creative Studio
- FLUX 1.1 Pro image generation (via Replicate)
- Claude-powered ad copy in 30+ languages
- Video generation from images/text
- Creative performance predictor (Claude Vision + 3-layer ML)
- Creative fatigue detection with 3-signal severity classification
- Winner detection via Bayesian A/B + Thompson Sampling auto-scale
Campaign & Automation
- Full CRUD on all 10 ad platforms (unlike Keitaro which only tracks)
- 14-operator rules engine with IF/ELSE logic
- Budget pacing with 5-check monitor and auto-cap
- Bulk campaign creation with CSV import
- Custom + lookalike audience builder
- Audience push to Meta, Google, TikTok audiences
Tracking & Intelligence
- 7-layer identity resolution + CNAME tracking (exceeds Keitaro)
- Server-side pixel + Triple Pixel compatibility
- Platform feedback via Meta CAPI, Google Enhanced Conversions, TikTok Events API
- Anomaly detection: Z-score + IQR + velocity analysis
- Competitor ad tracking with spend estimates
- Consent Mode v2 compliant (4 Google signals)
Scale & Scope
The Impact
Tech Stack
“Media buyers should not be human ETL pipelines. VelaReach ships 10 ad platforms, 11 attribution models, 14 AI features, and 9 microservices in one platform — replacing Keitaro, Dolphin, FabFunnel, Triple Whale, and Northbeam simultaneously. At $99-299/mo against Northbeam's $1K-15K/mo, it is 10x cheaper with 10x the feature surface. The lesson: microservices architecture lets you compete against five incumbents at once without any one service becoming a bottleneck.”