Aries AIBook a call

Strategy and roadmap · Architecture and implementation · Model training and deployment · System monitoring and evaluation

Your first AI systems, built by a world-class team.

We find where AI pays off in your business, build it into the tools your team already uses, and run it in production. Our people shipped AI features used by millions at Microsoft, built AI agents that manage $200m for an investment firm, and prediction systems that prevent millions of fraudulent messages in telecoms.

Take the 5-minute AI readiness check→Or book a call
Microsoft AIWindows and Copilot+ PC features
sanbo.ioAries client · $200m+ AUM
Senior teamNo juniors on any project
Model-independentOpenAI, Anthropic, Google, open-weight models

We bring experience from:

University of Cambridge
University of Oxford
Microsoft
Tenstorrent
Telesign
UNDP
X& Financial
PolyAI

Selected work

Where we work:
six areas.

Discuss your case →
Problem
Investment work spans conversations, documents and tools. A useful agent needs to retain context, learn the firm's way of working and complete workflows, not just answer isolated questions.
What we did
A Claude-powered managed agent, built cloud-hosted and multi-user with a proprietary internal harness, custom tools, an MCP layer and a context engine. Users work with it through Slack and sanbo.io. Model-provider processing meets zero data retention requirements, while a separate firm-controlled memory layer remembers previous conversations and skills taught by users. Those skills range from reproducing a branded report template to running end-to-end startup investment research, including benchmarking.
For your company
An agent can work across your team's tools, remember prior work and learn your processes, with a clear separation between internal memory and what model providers retain.

Result

  • Slack + sanbo.io
    shared cloud agent
  • Zero retention
    for model-provider data processing
  • Persistent memory
    conversations and user-taught skills
Read the X& Financial case study →

What our clients say

“We started with no AI team and no AI product. Eighteen months later sanbo.io runs more than $200m across six funds, and we scaled our deal flow 3x without adding more analysts. Luka built it, hired the people to run it, and then handed it over.”

Tiho Bajic

Managing Partner, X& Financial

Read the X& Financial case study

Engagements

How we work.

I.2 days

Executive Workshops

Two days to turn a list of AI ideas into a ranked plan with costs, which your leadership team can agree on.

You leave with a scored list of use cases, a cost and return range for the top three, and a 90-day plan.

  • ·Use-case discovery and scoring
  • ·Risk, governance and data review
  • ·Roadmap with metrics
II.8–24 weeks

Build & Deployment

We design, build and harden AI systems for production, such as copilots, document processing, scoring and agents, connected to the tools you already run. We can also train small custom models and run them on your own servers or in a private cloud, so your data stays with you.

  • ·Proof of concept, MVP, production
  • ·Small custom models trained on your data
  • ·Deployed on-prem, in a private cloud or hybrid
  • ·Evaluation, monitoring and guardrails
III.3–6 weeks

Focused Sprints

A fixed-length engagement to fix one specific bottleneck, run by a senior team.

  • ·3 to 6 weeks
  • ·Working demo in the final week
  • ·Pass/fail criteria agreed up front
IV.Quarterly engagement

Fractional Chief of AI

A senior AI leader embedded in your team one to three days a week. They set strategy, choose vendors and hire your first AI engineers.

  • ·Quarterly contracts
  • ·Hiring and team setup
  • ·Hands-on architecture review

In the media

The New York TimesWorld Economic ForumBloomberg AdriaBpifranceDataconomyEuronewsRTSPolitika

AI Readiness & Return Check

Quick check · indicative estimate

Find out how ready your company is for AI and what it could return.

In about five minutes you get a readiness score across five areas and a rough estimate of what an AI programme would cost and return.

Start the assessmentSee an example result