AI in Maysano

Maysano uses an LLM-powered assistant to support teams in creating, understanding, and governing data products aligned with the ODPS standard. The assistant combines conversational guidance with structured automation capabilities.

What AI Does

The AI supports multiple workflows across the platform:

Assistant Chat

Users can interact conversationally with the AI to:

  • Clarify governance concepts
  • Understand ODPS structures
  • Explore best practices
  • Get guidance on building data products

The assistant has contextual knowledge of ODPS and platform patterns to provide structured and consistent guidance.

ODPS Specification Generation

Users can describe a data product in natural language, and the AI generates a structured YAML specification aligned with the ODPS standard.

Generated specifications:

  • Follow structural templates
  • Include required fields
  • Respect validation rules
  • Are designed for further refinement by users

Specification Validation

The AI can review existing specifications and:

  • Identify missing fields
  • Detect structural inconsistencies
  • Highlight potential governance gaps
  • Suggest improvements

The goal is not replacement of governance review, but acceleration and support.

Governance Component Suggestions

Based on the type and context of a data product, the AI can recommend: - SLA profiles - Data quality profiles - Access profiles - Reusable governance components

Suggestions are aligned with reusable governance patterns.

Strategy Support

The AI helps teams articulate: - Product objectives - Outcome KPIs - Internal KPIs - Positioning and value propositions

This supports alignment between governance, product, and business roles.

Schema Assistance (Optional Integrations)

When integrated with metadata or catalog systems, the AI can:

  • Propose data schemas
  • Assist with schema mapping
  • Support column matching tasks

Executive Summaries & Insights

The AI can generate:

  • Executive summaries
  • Product overviews
  • Structured insights
  • Proactive analytical commentary

These outputs are designed for stakeholder communication and decision support.

How It Works (High-Level)

At a high level, Maysano uses an orchestration layer that:

  1. Receives a user request.
  2. Determines whether additional context is required.
  3. Optionally calls internal tools to fetch or update data.
  4. Produces a structured response streamed back to the user interface.

The system supports tool-enabled reasoning, allowing the assistant to both read and write within defined governance boundaries.

Safety & Guardrails

To improve reliability and predictability, Maysano applies multiple layers of controls:

  • Input constraints to prevent oversized or malformed requests
  • Timeout limits to prevent stalled responses
  • Token and cost budgets per feature
  • Failure handling mechanisms to avoid repeated retries during upstream degradation
  • Observability metrics for usage, latency, and cost tracking

AI suggestions are designed to support decision-making.
Final accountability always remains with human users.

Architecture Overview

flowchart TD subgraph FE["Frontend"] U[User] --> UI[AI Assistant UI] U --> Builder[Product Builder] U --> Validator[Specification Validator] U --> Mapper[Schema & Mapping] end subgraph API["Backend API"] UI --> A1["AI Orchestrator (streaming)"] Builder --> A2["Generate spec / strategy"] Validator --> A3["Validate specification"] Mapper --> A4["Schema helper / mapping"] end subgraph ORCH["AI Orchestration Layer"] A1 --> Loop[Tool-enabled reasoning loop] A2 --> Loop A3 --> Loop A4 --> Loop end subgraph SAFE["Safety & Guardrails"] Loop --> IV[Input validation] IV --> TB[Token & cost budgets] TB --> TO[Timeouts] TO --> FH[Failure handling] end subgraph LLM["LLM Provider"] FH --> GW[LLM gateway / provider] GW --> M[LLM model] end subgraph KB["Knowledge & Rules"] R["ODPS reference + templates + validation rules"] R -.-> Loop end subgraph TOOLS["Internal Tools"] direction LR Read["Read context:\nproducts\nprofiles\nversions\nartifacts"] Write["Write actions:\ncreate/update\nprofiles\nartifacts"] end Loop --> Read Loop --> Write Read --> DB[(Data Store)] Write --> DB subgraph OBS["Observability"] MET[Usage, cost, latency metrics] Loop -.-> MET end