Rachna Srivastava is an enterprise architect at the California Department of Financial Protection and Innovation who designs AI systems for financial-fraud detection and consumer protection. Her signature project, Cognitive Shield, combines specialized AI agents, knowledge graphs, and investigator-facing tools to identify synthetic identities, deepfake impersonation, voice-cloning scams, and cryptocurrency fraud.
Also professionally known as Rachana Srivastava, she has more than two decades of experience across enterprise software, distributed systems, cloud computing, and applied AI. Before joining California’s financial regulator, she worked at Synopsys, Ayla Networks, Hewlett Packard Enterprise, and Thomson Reuters. She credits Jeremy Howard and fast.ai with shaping her practical understanding of deep learning and commitment to accessible, ethical AI.
- Financial fraud is a network problem. Her Cognitive Shield demonstration shows how accounts, devices, identities, and transactions can expose coordinated fraud when analyzed together. Agentic workflows extract relationships from documents and messages; PostgreSQL stores operational information, while Neo4j maintains persistent fraud graphs.
- Investigators need usable graph intelligence. Through GraphRAG and natural-language-to-Cypher workflows, investigators can question graph data conversationally instead of writing specialized queries. Dashboards, automatic case escalation, audit trails, and compliance reporting connect detection to practical regulatory action.
- Different threats demand different agents. Her multi-agent fraud defense architecture assigns specialized components to threats including phishing, deepfakes, suspicious payments, and cryptocurrency scams. Streamlit, FastAPI, CrewAI, and LangChain support a modular system whose outputs remain subject to human review and monitoring for false positives and negatives.
- Sensitive AI requires controlled infrastructure. Her work on air-gapped public-sector AI addresses how agencies can analyze protected consumer information without surrendering control of sensitive data. Encryption, role-based access, explainability, governed data flows, and human oversight are foundational requirements.
