Vespa at Work:

Metal AI: Agent-driven intelligence for private equity, built on Vespa Cloud

Metal AI built an agent-driven intelligence platform on Vespa Cloud to automate due diligence and unlock faster, more consistent insights across complex, interconnected data.

At a Glance

  • ~95% of retrieval handled by AI agents
  • Faster due diligence workflows
  • More consistent, reusable outputs
  • Improved answer quality through context-aware retrieval

Introduction

Metal AI is an institutional intelligence platform built for private equity firms, enabling teams to extract, structure, and act on complex financial and operational data.

To power this vision, Metal AI built an agent-driven intelligence system on Vespa Cloud, where AI agents—not humans—handle the majority of retrieval, reasoning, and workflow execution.

The Challenge: Understanding Context with Fragmented, Highly Relational Data

Private equity workflows rely on fragmented, highly relational data:

  • Companies, people, deals, and documents are deeply interconnected
  • Critical insights are buried across PDFs, emails, and structured systems
  • Traditional search systems flatten relationships into keyword or vector matches

This creates major bottlenecks:

  • Time-consuming due diligence processes (e.g., DDQs)
  • Inconsistent answers across teams
  • Heavy reliance on manual synthesis

Metal AI needed a system that could:

  • Understand entities and relationships, not just text
  • Support multi-step reasoning workflows
  • Operate in real time at scale

The Solution

Metal AI built its retrieval and intelligence layer on Vespa Cloud, enabling a fully agent-driven architecture.

Vespa Cloud enables Metal AI to scale seamlessly:

  • No need to manage infrastructure
  • Handles large-scale, multi-entity datasets
  • Supports real-time query and update workloads

This allows the team to focus on building intelligence, not operating systems.

Key capabilities:

1. Multi-entity data modeling
Vespa allows Metal AI to represent companies, people, documents, and relationships as first-class objects—enabling agents to retrieve context-rich results.

2. Advanced ranking and retrieval
Instead of simple vector similarity, agents use Vespa’s ranking capabilities to combine:

  • Structured filters
  • Semantic signals
  • Business logic

3. Real-time filtering and updates
Agents operate on continuously updated data, ensuring outputs reflect the latest available information.

4. Agent-first architecture
AI agents handle ~95% of retrieval tasks, orchestrating complex workflows end-to-end.

Results: Automation, Accuracy, and Consistency

By building on Vespa Cloud, Metal AI achieved:

  • Massive automation: ~95% of retrieval handled by AI agents
  • Faster workflows: DDQs and similar processes completed significantly faster
  • Higher consistency: Standardized, reusable answers across teams
  • Improved accuracy: Context-aware retrieval grounded in real data relationships

What’s Next

Metal AI is building toward a future where:

  • AI agents orchestrate entire investment workflows
  • Retrieval, reasoning, and execution are fully integrated
  • Human users focus on oversight and decision-making

Vespa provides the foundation for this shift, from search systems to agent-driven intelligence platforms.

More Reading

Autoscaling with Vespa

This eBook explores how Vespa’s advanced autoscaling capabilities help organizations efficiently manage variable workloads by automatically adjusting resources to meet performance, cost, and scalability requirements.

Migrating from Elasticsearch to Vespa

In this webinar, guest speaker Ravindra Harige, long-time search expert and founder of Searchplex, will share how to confidently make the move from Elasticsearch to Vespa and scale with confidence.

GigaOm Radar for Vector Databases V3

This report provides a detailed comparison of 17 leading open source and commercial solutions, examining their strengths across hybrid search, semantic retrieval, RAG, and large-scale AI workloads.