Enterprise AI Platform

About:

This enterprise-grade tool-based agentic AI platform achieved 50%+ adoption with nearly 100 daily active users at a mid-sized software company. Starting in 2023, I architected and led development of a platform with 10+ specialized AI tools enabling autonomous multi-step workflows across database querying, RAG search, data analysis, and visualization.

The platform serves as the central AI infrastructure for the organization, integrating with existing business systems (ERP/CRM, support ticketing, documentation, KB, analytics databases) to provide contextual intelligence, automation, and measurable business impact including proven ARR growth through SaaS upsells.

Platform Architecture:

Tool-Based Agentic System:

• 10+ specialized AI tools enabling autonomous multi-step workflows
• Non-trivial cross-tool data piping (passing datasets between SQL → Python → visualization)
• 50% token reduction through JSON-to-CSV data transformation
• DuckDB integration for efficient ERP query execution

Specialized Tool Suite:

• SQL query execution on PostgreSQL analytics databases
• ERP saved search integration with DuckDB-powered query layer
• Multi-domain RAG search (KB articles, product docs, 70,000+ support cases, HR documents)
• Sandboxed Python evaluation for math and statistical operations
• Dynamic data visualization (line/bar/scatter charts generated from tool outputs)

Enterprise Integration:

• ERP/CRM with MCP connector
• PostgreSQL analytics databases
• Apache Superset for reporting
• Microsoft Entra SSO authentication
• Real-time data pipelines across multiple business systems

Advanced Optimization:

• Custom hybrid search combining vector embeddings with LLM-engineered metadata
• Production-scale token management and quota optimization
• Efficient data transformation and caching strategies

Deployed Applications:

Business Intelligence & Analytics:

• Natural language queries replacing manual Excel workflows
• Complex search capabilities (“find customers in Florida in healthcare vertical shipping 10k+ packages/week”)
• Financial analytics (“show all reps, calculate ARR, accounts, ARR/account, key metrics”)
• Saves sales management 50 hours/quarter on one query type alone
• Support teams ask questions → system autonomously searches KB + docs + 70,000+ support cases

Proactive Sales Intelligence Suite:

• Six autonomous alert workflows monitoring customer usage patterns
• WMS license overuse detection (proven ARR impact from upsells)
• Shipping label consumption forecasting
• License underutilization monitoring (downsell risk)
• Usage drop-off detection for rapid response
• Managed services optimization with tailored recommendations
• Advanced functionality adoption tracking
• Slack-integrated delivery to sales team
• Real-time customer data processing with pattern recognition

AI Support Assistant:

• RAG architecture processing 70,000+ historical support cases
• Transforms helpdesk from reactive ticket creation to proactive case prevention
• Recognized as “biggest productivity enhancer we’ve ever had” by 17-year company veteran

Process Automation:

• Sales-to-services transition automation (saves 2 hours per customer handoff)
• Technical modification search and comparison
• Product customization knowledge base
• Rapid identification of similar past implementations

Technologies Used:

• Backend: Python, FastAPI, Streamlit
• AI/LLM: AWS Bedrock, Claude Sonnet, LangChain
• Databases: Milvus (vector), PostgreSQL, DuckDB (data transformation)
• Cloud Infrastructure: AWS (ECS, RDS, EC2, Bedrock), Terraform (IaC), Docker, GitHub Actions CI/CD
• Enterprise Integrations: ERP/CRM (MCP connector), Microsoft Dynamics 365, Apache Superset, Entra SSO
• Data Pipeline: ERP API, PostgreSQL analytics, multi-system data orchestration

Leadership & Advocacy:

• Lead monthly “AI for All” training workshops for cross-departmental AI literacy
• Regular presentations to C-suite and director-level stakeholders
• Technical roadmap management for platform expansion
• Mentor junior developers on AI development practices
• Translate complex AI capabilities into strategic business value propositions

Technical Achievements:

• Architected tool-based agentic platform achieving 50%+ adoption at mid-sized company
• Engineered 50% token reduction through JSON-to-CSV data transformation
• Designed non-trivial cross-tool data piping system for complex workflows
• Built custom hybrid search combining vector embeddings with LLM-engineered metadata
• Oversaw infrastructure migration to production AWS with Terraform IaC and CI/CD
• Implemented DuckDB for efficient ERP query execution
• Created sandboxed Python evaluation environment with statistical analysis capabilities
• Delivered measurable business impact: proven ARR growth, 50 hrs/quarter time savings, 2 hrs/customer automation