Srivastava frames trust as an engineering property rather than a policy promise. Financial investigations can reach court years after a system produced a result, so every step must remain explainable, reproducible and auditable while sensitive consumer information stays protected.
The architecture separates data ingestion, processing and reasoning instead of treating one language model as a magic box. Kafka buffers uneven fraud traffic and keeps events in order, Spark cleans and transforms inputs, and the language model handles reasoning. Personally identifiable informationPersonally identifiable information is data that directly identifies a person or can reasonably be combined with other data to identify them. is redacted before the reasoning layer, while cryptographic controls bind protected data to hardware in the server rack.
Limited offline computeAir-gapped AI runs inside an isolated environment with no direct network path to less trusted systems or the public internet. makes routing essential. A semantic routerA semantic router classifies the meaning or intent of an input and directs it to the model, tool or workflow best suited to handle it. sends each task to the smallest capable model, allowing routine summarization and extraction to avoid an unnecessarily large model. Srivastava says this design handled substantially more traffic without adding GPUs while reducing processing cost.
A one-way optical data diodeA data diode is a hardware-enforced one-way communication device that permits data to flow in only one direction between security zones. lets new threat information enter without creating a return path for sensitive data. Incoming material first lands in quarantine for validation, and accepted records are written to an immutable, time-travel-capable storeAn immutable data store preserves prior records and represents changes by appending new entries instead of rewriting history. so investigators can reconstruct the state that produced an earlier decision. The talk's broader claim is that hardware boundaries, data discipline and reproducibility must be designed together for high-stakes AI.
Watch on YouTube
