OddsAgg
A real-time sports betting odds aggregation platform, built from the ground up as a distributed microservices system processing data from 50+ bookmakers across 30+ sports.
Overview
OddsAgg aggregates real-time betting odds from over 50 bookmakers and prediction markets worldwide, processing more than 3 million odds updates every day. The system normalises data from disparate sources into a unified schema, calculates consensus pricing across all providers, detects arbitrage opportunities in real-time, and streams updates to connected clients via WebSockets.
The platform is built as a NestJS microservices monorepo with complete TypeScript coverage. It employs event-driven architecture with BullMQ job queues and Redis pub/sub, multi-database persistence via Drizzle ORM, Kubernetes orchestration with horizontal pod auto-scaling, and a full observability stack spanning Prometheus, Grafana, and the ELK suite.
Data sources include traditional sportsbooks via REST API polling, private APIs, premium data providers via RabbitMQ, and decentralised prediction markets via protocol-specific readers. Each feed is abstracted behind a unified interface with self-registering service discovery and automatic health monitoring.
The aggregated data is exposed to customers through a commercial API available over both HTTP and WebSocket, with binary delta updates to minimise payload sizes and deliver optimal real-time updates. Access is tiered across Pro, Business, and Enterprise plans.
Delivery
The project ran from September 2025 to January 2026. We worked weekly with the client, embedded directly with the product team. Architecture decisions were made collaboratively, with Amplir owning the technical execution end to end.
Delivery was phased: core infrastructure, the data pipeline, and Kubernetes orchestration shipped first. The consensus engine, arbitrage detection, API layer, and dashboard followed in the second phase. The observability stack and production hardening came next. Feed integrations were the final phase, building on the unified ingestion interface to connect each source with its own normalisation logic, error handling, and health reporting. Every phase shipped to a live Kubernetes environment with full CI/CD from day one.
Amplir integrates AI-assisted development tooling into every stage of delivery. Combined with experienced engineers who know what to build and how to architect it, this allows a small team to move at a pace that would traditionally require a significantly larger team over a longer timeline. A system of this scope and complexity would typically represent a seven-figure investment at standard enterprise consultancy rates. Amplir delivered it in three months of active development at a fraction of that.
System Architecture
A layered microservices architecture with clear separation between data ingestion, processing, storage, and delivery.
Technology
The backend is built entirely in TypeScript on NestJS, with Node.js as the runtime. Data persistence uses PostgreSQL 15 across three databases (events and odds, authentication, and configuration), all managed through Drizzle ORM with type-safe migrations and Zod-based runtime validation.
The messaging layer combines BullMQ for reliable job queue processing, RabbitMQ for consuming premium data feeds, and Redis pub/sub for broadcasting real-time odds updates across services. Client-facing real-time delivery uses Socket.IO WebSocket gateways with channel-based subscriptions.
Data ingestion strategies include REST API polling for standard bookmaker integrations, private APIs, and protocol-specific readers for decentralised prediction markets.
Infrastructure runs on Kubernetes with horizontal pod auto-scaling, fronted by a Kong API Gateway for rate limiting and request transformation. The CI/CD pipeline is automated through GitLab with multi-stage Docker builds pushed to a private registry. API authentication uses JWT via Passport.js with role-based access control, while internal tooling and dashboards are gated behind Authelia as an SSO proxy. Audit logging, Kubernetes network policies, and non-root containers round out the security posture.
Key Features
A closer look at the core capabilities of the platform.
Multi-Strategy Feed Ingestion
Over 50 data sources are connected through a single integration layer, regardless of how each provider exposes their data. Public APIs, private APIs, premium data feeds, and decentralised prediction markets are all supported out of the box.
New sources can be added without changes to the core platform. Each connection is self-managing: it registers automatically on startup, reports its own health status, and is removed from the active pool if it becomes unresponsive. The result is a system that scales to new providers with minimal operational overhead.
Queue-Based Data Pipeline
Every piece of incoming data is processed through a managed pipeline that guarantees delivery. Critical structural data (events, teams, leagues) is always processed ahead of odds updates, so the platform never serves stale or incomplete context.
No single data source can overwhelm the system. Built-in rate controls operate at both the individual source and platform-wide level. All incoming formats are automatically converted into a single consistent structure, and duplicate records are eliminated before they ever reach the database. The result: over 3 million updates processed daily with at-least-once delivery guarantees and no observed data loss in production.
Statistical Consensus Engine
Rather than showing raw odds from every bookmaker, the platform produces a single consensus view of where the market sits for any given event. Users can instantly see whether a bookmaker is pricing above or below the market, and by how much.
Different source types (traditional bookmakers, prediction markets, and exchanges) are analysed independently so that fundamentally different pricing models don't distort the picture. Every consensus price carries a confidence score so users know how much weight to give it based on how many sources contributed, how recent the data is, and how reliable those sources have been historically.
Real-Time Arbitrage Detection
The platform continuously monitors odds across all connected bookmakers and flags guaranteed-profit opportunities the moment they appear. When pricing discrepancies create a window where backing all outcomes across different bookmakers locks in a profit regardless of result, users are notified instantly.
Each detected opportunity includes the exact profit margin, proportional stake allocation across outcomes, and a confidence rating based on data quality. Alerts are delivered in real time to dashboards and notification channels, giving users the information they need to act before the market corrects.
Prediction Market Comparative Analysis
Traditional bookmakers and prediction markets price risk differently. Bookmakers set odds based on liability management and margin targets, while prediction markets reflect crowd-sourced probability estimates driven by real capital. Significant movements in either can signal information the other hasn't yet priced in.
The platform runs continuous comparative analysis between bookmaker consensus and prediction market pricing. When the two diverge beyond configurable thresholds, the system flags the discrepancy and surfaces it alongside historical convergence data. This cross-source intelligence gives users an edge that no single bookmaker or prediction market can provide on its own.
Commercial Data API
All aggregated odds, consensus pricing, and arbitrage data is available through a commercial API. Clients can pull data on demand or subscribe to a real-time feed that pushes only what has changed, keeping connected systems synchronised with minimal bandwidth and latency.
Access is structured across three tiers (Pro, Business, and Enterprise), each with its own rate limits, data scope, and support levels. Usage is metered per API key, giving full visibility into consumption and straightforward billing for downstream customers.
Real-Time Dashboard
A live web dashboard gives users a complete view of odds across all connected bookmakers and prediction markets, updated in real time as data arrives. Users can browse by sport, event, or market and instantly see consensus pricing, individual bookmaker odds, and historical movement charts.
Arbitrage opportunities, prediction market divergences, and market movement indicators are surfaced automatically. Filtering, sorting, and watchlist features let users focus on the markets that matter to them. The interface is designed for speed and density, prioritising data clarity over decoration.
Observability & Operations
Every service, database, and queue in the platform is continuously monitored. Operational dashboards provide a live view of system health, feed performance, and data quality across the entire infrastructure. If something degrades, the team knows before it impacts users.
Centralised logging captures every request across all services with full traceability, making it possible to follow a single data point from ingestion to delivery. Automated alerting notifies the operations team through multiple channels when key thresholds are breached, ensuring rapid response times and minimal downtime.
Infrastructure & Deployment
The platform scales automatically to meet demand. Under heavy load, additional capacity is provisioned within seconds and scaled back down when traffic subsides. Every service is health-checked continuously, and unhealthy instances are replaced without manual intervention.
Deployments are fully automated with zero downtime. Code changes move through automated testing and validation before reaching production, and every release rolls out gradually with automatic rollback if issues are detected. The infrastructure is designed to be hands-off, letting the engineering team focus on features rather than operations.
Results
The platform was taken from initial concept to production deployment with three months of active development, delivering a fully operational system with over 80 deployed services (including 50+ independent feed connectors) processing more than 3 million daily odds updates across 30+ sports.
Every commit is automatically linted, tested, built, and deployed to Kubernetes through a fully automated CI/CD pipeline. The observability stack provides complete operational visibility. Every metric, log, and alert is centralised and actionable. The security posture includes JWT authentication with role-based access control, Kubernetes network policies, non-root containers with dropped capabilities, and full audit logging.
The system is horizontally scalable by design, with auto-scaling pods that handle traffic spikes without manual intervention, and a queue-based architecture that gracefully absorbs load through backpressure rather than dropping data.
“We gave Amplir a big ask and they delivered. Three months of actual build time, 80+ services running in production, full observability, a commercial API. It's rare to find a team that operates at this level without needing to be managed. We're continuing to build with them.”
What's Next
The platform is actively evolving. Here's what we're building together.
AI-Powered Betting Recommendations. With millions of historical odds movements and outcome data accumulating daily, the next phase introduces machine learning models trained on this dataset to surface value bets, identify mispriced markets, and generate confidence-scored recommendations.
Market Movement Alerts. Real-time notifications when significant line movements, sharp money indicators, or prediction market divergences are detected. Configurable per sport, market type, and threshold, delivered via push notification, email, or WebSocket.
Automated Betting Execution. Closing the loop from insight to action. The system will support automated bet placement on prediction markets via their public APIs, driven by user-defined strategies with risk controls, bankroll management, and full audit trails built in from the start.
Cross-Market Data Intelligence. The platform's growing dataset of historical odds movements, outcome data, and market behaviour extends well beyond traditional sports. The next phase applies AI-driven analysis to adjacent markets including political elections, entertainment awards, and economic events, identifying pricing patterns, sentiment shifts, and cross-market correlations that manual analysis would miss. The goal is a general-purpose prediction intelligence layer built on top of the aggregation engine.
Have a project like this?
Whether it's real-time data platforms, cloud infrastructure, or AI-integrated products, we deliver enterprise-quality software, end to end.
Get in touch