Observing a sky observator, Making app faster and Automated Reconciliation
⚡ How Agoda Achieved 6x Faster Time-Series Queries with TimescaleDB.
What happened – Agoda switched its analytics portal to TimescaleDB, slashing P95 query latency 6× and eradicating timeouts.
Why it matters – Shows TimescaleDB can turbo-charge heavy time-series workloads on Postgres with minimal code change.
🤖 How GitHub Copilot Serves 400 Million Completion Requests a Day
What happened – GitHub revealed Copilot’s global GPU fleet, batching layer and HTTP/2 mux handling 400 M daily completions under 100 ms.
Why it matters – Offers a concrete blueprint for scaling generative-AI inference APIs efficiently and reliably.
🏠 Embedding-Based Retrieval for Airbnb Search
What happened – Airbnb deployed vector-embedding retrieval backed by Faiss, boosting search recall and booking conversion.
Why it matters – Demonstrates adding low-latency vector search over billions of items without exploding compute spend.
💵 Building Pipelines for Automated Reconciliation with Near-Zero Data Loss
What happened – Walmart built a Kafka-Flink pipeline that replays missed events and reconciles payouts automatically, cutting manual effort and losses.
Why it matters – Provides an exactly-once streaming pattern for financial correctness at multi-billion-event scale.
🔭 Skyscanner’s Journey to Effective Observability
What happened – unified logs, metrics and traces on OpenTelemetry with SLO dashboards, reducing issue detection time by 70 %.
Why it matters – Roadmap for consolidating observability tooling to raise reliability without ballooning costs.
📱 Supercharging Discord Mobile: Our Journey to a Faster App
What happened – new architecture, new image pipeline result—cutting cold-start time 43%.
Why it matters – Gives cross-platform teams actionable tactics for shrinking bundle size and bridge overhead.
🐉 FeatureStore at Agoda: Optimizing Dragonfly for High-Performance Caching
What happened – Agoda tuned Dragonfly’s sharding and pipelines to serve millions of feature QPS with one-third the nodes of Redis.
Why it matters – Shows modern in-memory stores can cut ML serving costs while staying API-compatible with Redis.
If you are into GO then this article from Uber on their experience with profile-guided optimizations in GO
Got a link that belongs here, or any feedback? Reach out to me on LinkedIn, and I’ll check it out.
Until next time – stay scalable! ✌️