Building Agentic AI Applications with PostgreSQL as the Backbone
Presented by:
Shayon Sanyal
Shayon is a database architect specializing in PostgreSQL and AI/ML applications. With extensive experience helping enterprises implement vector search, agentic AI systems, and hybrid search architectures, he focuses on practical patterns that work at production scale. A regular speaker at database conferences, Shayon is passionate about making PostgreSQL's advanced capabilities accessible to the broader community.
No video of the event yet, sorry!
The AI landscape is shifting from chatbots to autonomous agents—systems that plan, use tools, maintain memory, and take actions. While attention focuses on language models, the real differentiator for production agents is the data layer. PostgreSQL, with its combination of relational integrity, vector search, and extensibility, is emerging as the ideal backbone for agentic AI.
This session explores architectural patterns for building production-grade AI agents with PostgreSQL at the core.
Memory Architecture: Agents need both short-term conversation context and long-term memory.
We'll cover implementing:
- Episodic memory: conversation history with efficient retrieval
- Semantic memory: knowledge retrieval via pgvector similarity search
- Procedural memory: learned patterns stored as retrievable examples
All in PostgreSQL with proper isolation and retrieval patterns.
Tool and Function Registries: Agents invoke tools to take actions. We'll demonstrate:
- Storing tool definitions and parameter schemas
- Dynamic tool discovery based on context
- Execution metadata and audit trails
- Approval workflows for sensitive operations
Model Context Protocol (MCP) Integration: MCP is becoming the standard for connecting AI assistants to external systems. We'll show how to integrate MCP servers that expose PostgreSQL capabilities:
- Schema introspection for natural language queries
- Safe, parameterized query execution
- Structured data retrieval with proper formatting
State Management: Complex agents maintain state across multi-step tasks:
- Workflow state persistence
- Checkpoint and resume patterns
- Reliable task queues for agent orchestration
- Handling failures and retries gracefully
Grounding and Guardrails: Agents hallucinate. We'll demonstrate:
- Grounding responses in PostgreSQL data
- Fact-checking queries before responding
- Safety guardrails preventing harmful actions
- Confidence scoring based on data availability
Live demonstrations will show a working agent system that queries databases, maintains conversation context, and takes actions—all orchestrated through PostgreSQL. We'll discuss why PostgreSQL beats purpose-built vector databases for agent workloads: ACID transactions for reliable state, mature tooling, and the ability to join vector results with structured data in a single query.
Key Takeaways:
- Design memory systems giving agents persistent, queryable context
- Implement tool registries with safety controls
- Integrate MCP servers for PostgreSQL
- Apply guardrails keeping agents grounded in real data
- Date:
- Duration:
- 50 min
- Room:
- Conference:
- Postgres Conference: 2026
- Language:
- Track:
- Dev
- Difficulty:
- Medium