At Doppel, we build machine learning systems to detect and disrupt social engineering attacks. The threat landscape changes constantly, which means our models need to evolve just as quickly.
That translates into two core requirements for our ML workloads:
- Fast experimentation
- Reliable and scalable inference
For a long time, infrastructure friction slowed us down in both places.
On the training side, the biggest bottleneck was experimentation throughput. Experiments often ran sequentially, forcing new ideas to wait for previous runs to finish and slowing down iteration. With Modal, we can run experiments in parallel using simple Python abstractions, allowing us to evaluate many hypotheses at once and significantly shorten the feedback loop between ideas and results.
On the inference side, our previous setup required packaging models into Docker containers, deploying GPU-backed services, and maintaining HTTP endpoints. This introduced operational overhead and slowed down iteration. Modal simplifies this by reducing build times, handling scaling automatically, and allowing inference to be invoked directly—removing much of the integration code we previously maintained.
Across both training and inference, the biggest change was reducing the operational costs around our models—making it easier to iterate quickly and ship detection systems. You can read the full post on Modal’s blog here.
About Doppel Engineering
Doppel is building the leading AI-native social engineering defense platform, protecting organizations from the fastest-growing threat in cybersecurity: AI-powered impersonation and deception. Our engineering team operates at the intersection of AI, large-scale threat intelligence, and real-time security infrastructure. We build systems that detect and dismantle attacker campaigns across the internet while strengthening human resilience against social engineering attacks. This blog is where we share how we build, from LLM-driven detection pipelines and autonomous takedown systems to human risk management technologies, and the engineering challenges behind defending the modern internet from social engineering at scale.
