Fisher Investments' Enterprise AI Architecture represents a comprehensive, battle-tested solution built on open source components and Azure cloud infrastructure. Managing $362 billion in assets under management (AUM) across 174,538 client accounts, this architecture seamlessly integrates data from Salesforce Financial Services Cloud, Bloomberg Terminal, ServiceNow, Workday, and proprietary knowledge systems to power intelligent decision-making across the organization.
The platform leverages NVIDIA-accelerated AI/ML orchestration with cutting-edge capabilities including model serving (NVIDIA Blueprints, Model Registry), agent frameworks (AgentBlocks, Langgraph, MCP, A2A), real-time ML (NVIDIA Triton, KServe), large language model services (NVIDIA NIM, Llama Family, NVIDIA Nemotron), text-to-SQL generation (Databricks Genies, Databricks Deep Insights), MLOps (MLflow, Experiments, Model Versioning), vector search (Milvus GPU, NVIDIA CuVS), and feature stores (Feast, Drift Monitoring).
Built on a robust Databricks-powered data platform with Delta Lake architecture, the system processes real-time market data, client interactions, and operational metrics through a unified semantic layer powered by dbt and Databricks Unity Catalog. Azure Purview ensures comprehensive data governance with automated PII detection, lineage tracking across extensive datasets, and SEC/FINRA-ready audit reports.
The architecture delivers measurable business impact: the AI Agent Assist system handles significant annual inquiries with high resolution rates, Kitchen Sink application processes high weekly lead query volumes with conversion improvement, and AI-Powered Targeting achieves higher engagement through intelligent customer segmentation. The Fisher KI (Knowledge Intelligence) system serves as a centralized repository of investment philosophy and operational excellence. Security and compliance are paramount, with SOC 2 Type II certification, multi-factor authentication across all 6,300 employees, end-to-end encryption (AES-256 at rest, TLS 1.3 in transit), and high uptime SLA with robust disaster recovery.
Intelligent Client Service Automation
AI-powered assistant handling significant annual client inquiry volumes across phone, email, and chat channels. Leverages NVIDIA NIM and Llama models with real-time context from Salesforce FSC, reducing average handling time significantly.
Lead Qualification & Conversion Engine
Processes high weekly lead inquiry volumes through intelligent scoring and routing. Uses Databricks Genies for text-to-SQL querying across lead database, Bloomberg market data, and Salesforce opportunity pipeline.
Intelligent Marketing & Personalization
Advanced segmentation engine using Milvus vector search and NVIDIA CuVS to identify micro-segments across 174,538 clients. Drives personalized content, outreach timing, and product recommendations.
Knowledge Intelligence Repository
Centralized AI-powered knowledge system indexing Ken Fisher's books, investment philosophy, policy documents, and operational procedures. Powered by vector search (Milvus GPU) and retrieval-augmented generation (RAG) with NVIDIA NIM.
NVIDIA Blueprints provide pre-built AI workflows for financial services, while the Model Registry tracks versions, performance metrics, and deployment history across production models.
• NVIDIA Blueprints: pre-optimized AI workflows
• Model Registry: version control, lineage tracking
• A/B Testing: champion/challenger comparison
• Canary Deployment: gradual rollouts with monitoring
• Auto-scaling: dynamic resource allocation
AgentBlocks and Langgraph enable complex multi-agent workflows for AI Agent Assist and Kitchen Sink. MCP (Model Context Protocol) ensures consistent context sharing, while A2A (Agent-to-Agent) enables autonomous collaboration.
• AgentBlocks: modular agent components
• Langgraph: visual agent workflow design
• MCP: shared context across agent interactions
• A2A: autonomous agent-to-agent communication
• Tool integration: Salesforce, Bloomberg APIs
NVIDIA Triton Inference Server and KServe deliver low-latency model predictions for real-time applications. NVIDIA vLLM accelerates LLM inference for conversational AI with optimized GPU utilization.
• Triton: high-throughput predictions across A100 cluster
• KServe: Kubernetes-native serving with auto-scaling
• vLLM: significantly faster LLM inference vs. standard deployment
• Model ensembles: combine multiple models per request
• GPU sharing: multi-model deployment per GPU
NVIDIA NIM (NVIDIA Inference Microservices) provides optimized LLM deployment, while Llama models power conversational AI. NVIDIA Nemotron delivers financial domain expertise with fine-tuned models on Fisher data.
• NVIDIA NIM: containerized LLM deployment
• Llama 3.3 70B: primary conversational model
• Nemotron: financial domain specialization
• Fine-tuning: Fisher-specific vocabulary, tone
• Safety guardrails: content filtering, compliance checks
Databricks Genies converts natural language questions into optimized SQL queries, enabling business users to access data without technical expertise. Deep Insights provides automated analysis and recommendations.
• Natural language interface: "show me top advisors by AUM"
• Context-aware SQL: understands table relationships
• Deep Insights: automated trend detection, anomalies
• Query optimization: efficient execution plans
• Audit trail: tracks all user queries for compliance
MLflow orchestrates the complete ML lifecycle from experimentation to production deployment. Tracks extensive experiments with automated versioning, performance monitoring, and model governance.
• Experiment tracking: hyperparameters, metrics, artifacts
• Model registry: centralized version control
• Performance monitoring: accuracy, latency, drift detection
• Automated retraining: triggers based on performance thresholds
• Reproducibility: containerized environments per model
Milvus GPU-accelerated vector database powers semantic search across Fisher KI documents, client profiles, and market research. NVIDIA CuVS provides optimized similarity algorithms for A100/H100 GPUs.
• Millions of vectors indexed: documents, clients, securities
• Low query latency with GPU acceleration
• NVIDIA CuVS: optimized HNSW, IVF algorithms
• Hybrid search: combines semantic + keyword matching
• Real-time indexing: new content searchable within 1 min
Feast feature store provides consistent feature definitions across training and inference. Drift monitoring detects distribution changes that degrade model performance, triggering automated retraining.
• 500+ features: client demographics, portfolio metrics, behavior
• Real-time serving: low-latency feature retrieval
• Point-in-time correctness: prevents data leakage
• Drift detection: statistical tests on feature distributions
• Feature lineage: tracks transformations from raw data
Azure Event Hubs streams high-volume events daily from Bloomberg, Salesforce, ServiceNow. Azure Data Factory orchestrates extensive batch pipelines. Debezium captures database changes from Workday, legacy systems.
• Event Hubs: high daily message volume, low latency
• ADF: extensive pipelines, substantial data processed daily
• Debezium: CDC from Oracle, SQL Server, PostgreSQL
• Schema validation: ensures data quality at ingestion
• Dead letter queues: failed messages for retry
Azure Data Lake stores raw, bronze, silver, and gold data tiers (10+ PB). Azure SQL for structured reporting databases. Cosmos DB for low-latency, globally distributed application data.
• Data Lake: 10+ PB across hot, cool, archive tiers
• Azure SQL: multiple databases for reporting, analytics
• Cosmos DB: low read latency for apps
• Lifecycle policies: auto-tier data based on age
• Geo-replication: RA-GRS for disaster recovery
Azure Purview provides unified data governance, discovery, and compliance management across Fisher's entire data estate with automated PII detection.
• Data catalog: extensive datasets searchable
• Lineage: trace data from source to dashboard
• Sensitivity labels: auto-detect SSN, financial data
• Access analytics: who accessed what, when
• SEC/FINRA audit-ready reports
Kafka event streaming with high daily message volumes. Databricks Lakehouse processes substantial data daily with Delta Lake ACID transactions. Spark jobs transform raw data into analytics-ready gold tables.
• Kafka: high daily event volume, topic partitioning
• Databricks: extensive Spark clusters, auto-scaling
• Delta Lake: ACID, time travel, schema evolution
• Spark: batch + streaming, high daily throughput
• Medallion architecture: bronze, silver, gold layers
dbt + Databricks Unity Catalog - Unified business metrics ensuring consistency across all AI applications and analytics. Features include Client 360 Views, Advisor Performance Dashboards, and Regulatory Reporting with single-source-of-truth definitions.
Fisher's cloud-native infrastructure built on Microsoft Azure provides enterprise-grade compute, storage, and networking with multi-region deployment for high availability. NVIDIA GPU infrastructure powers AI/ML workloads.
• Primary: East US 2 (near Plano, TX HQ)
• Secondary: West US 2 (DR site)
• AKS: extensive nodes for AI workloads
• NVIDIA A100/H100 GPU pools
• High uptime SLA with auto-scaling
Multi-layered security architecture ensuring data protection, access control, and compliance with financial services regulations (SEC, FINRA, SOC 2).
• Azure AD: SSO + MFA for all 6,300 employees
• Key Vault: secrets, API keys, certificates
• RBAC: least-privilege access
• Encryption: AES-256 at rest, TLS 1.3 in transit
• Annual SOC 2 Type II audits
Private networking with ExpressRoute dedicated connections from Fisher offices to Azure, ensuring low-latency and secure data transmission.
• VNet isolation: application, data, gateway subnets
• Private Link: Azure services never on public internet
• ExpressRoute: high-bandwidth dedicated connection
• Azure Firewall: application-aware filtering
• DDoS Standard: automatic attack mitigation
Comprehensive monitoring and alerting across infrastructure, applications, and ML models with distributed tracing for debugging complex AI workflows.
• Azure Monitor: metrics, logs, alerts
• Application Insights: APM for AI apps
• Log Analytics: centralized log aggregation
• OpenTelemetry: distributed tracing end-to-end
• Custom dashboards: real-time KPIs
Ken Fisher Books
Policy Documents
SharePoint Docs
Client CRM
174,538 Accounts
Activity Tracking
IT Service Mgmt
Significant Annual Tickets
Change Tracking
Market Data
Research Reports
Live News Feeds
HR Management
6,300 Employees
Org Hierarchy
Enterprise AI Architecture Solutions