Engineering for the platforms behind programmatic
Appliscale is an adtech engineering company with eleven years building bidder, attribution and reporting infrastructure for global SSPs and DSPs. Our engineers have shipped platforms handling over 20 billion impressions per day, and bidders we built are live on exchange traffic right now.
We work on the hot path: auction logic, pacing, low-latency serving, and the data systems that make sense of what happened. Deep technical writing on this work lives on our dedicated vertical site at adtech.appliscale.io.
DSP and SSP engineering end to end
We built Bedrock Platform end to end, as both engineering partner and investor: a fault-tolerant containerized DSP handling 500,000 queries per second with sub-100ms responses, running inside IndexCloud, Index Exchange's proprietary datacenter. Our scope covered the bidder, infrastructure, VictoriaMetrics and Grafana observability, CI/CD and a cost-effective event aggregation pipeline for reporting.
At Yahoo we provided technical leadership on an RTB platform processing up to 1.5 million bid requests per second across hundreds of machines in four global regions. We migrated on-premise services to cloud-native Erlang and Java, added ML-optimised bidding that lifted revenue, and introduced CI/CD that let teams release multiple times a day.
Real-time bidding infrastructure that holds under load
Latency budgets in programmatic leave no room for a slow region or a cold cache. We design multi-region active-active systems on AWS and GCP: EU traffic terminates on AWS, APAC traffic on GCP, and every request is answered by the nearest region in under 100 milliseconds.
On Kubernetes we use EKS with Karpenter for demand-driven node scaling. On EC2 fleets we build dynamic scaling policies from traffic statistics; for one of the largest players in the US advertising market that approach, combined with immutable auto-healing infrastructure, cut EC2 costs by roughly 40% monthly on a platform serving billions of requests. For Spearad we built multi-region infrastructure with latency-based routing that can stand up a new region in days.
Header bidding data pipelines and reporting at auction scale
Auctions generate more data than they generate revenue, so pipeline economics decide margins. We built an ingestion and processing pipeline handling 400M+ daily header-bidding events and over 1 TB of data on ECS, Glue, Spark and Redshift, with autoscaling that follows traffic and near-realtime aggregation for reporting.
Cost work is part of the job. We took one platform's data lake spend from roughly $500K per month to $10K, a 98% cut, by reworking storage formats and matching query engines to workloads (Redshift versus Athena is a decision, not a default). See our data engineering practice for the full offering.
Beyond pipelines, we built a DeepAR-based forecasting engine that predicts campaign delivery so AdOps teams can schedule more campaigns without underdelivery risk, and a serverless cookieless geo-contextual targeting PoC that carried a startup through two successful investor roadshows in six months.
Where adtech goes next: agentic
Programmatic is moving from UI-driven workflows to agent-to-agent transactions, and we build for that transition today:
- Seller Agent development: agent-readable inventory, adagents.json and ad server booking for publishers and SSPs.
- Buyer Agent development: brief-to-booked buying for agencies and brands, on AdCP and AAMP.
- ARTF Containers: the IAB Tech Lab's Agentic Real Time Framework, service agents and bidders running inside the host platform's infrastructure, so data never leaves and egress costs drop. Bedrock, which our team launched, was the first containerized DSP.
- Prebid and Trusted Server: server-side Prebid setups and Trusted Server work on Fastly.
For protocol analysis, teardown posts and implementation notes, read the adtech vertical blog.
Talk to an engineer
