Two kinds of AI work, and we do both
AI in your product. We design and build AI capabilities your users and operators touch: agent architectures, documentation-grounded assistants, and automations that remove manual steps from a workflow.
AI in how we build. Our teams use AI throughout delivery: conversion tooling for large migrations, AI-generated test coverage, and delivery automations that keep a small team moving at the pace of a larger one.
The second makes the first cheaper and faster. Clients get the product and a delivery engine that keeps paying off.
What this looks like in practice
- A migration team built around AI tooling. For an enterprise software vendor we built a migration team that converts legacy Apex code to C# .NET and generates field mappings with tools, not by hand. One customer migration alone covered roughly 2,500 files of custom code. The method is written down as a reusable playbook the client's own engineers can run.
- A fintech platform delivered with AI in the loop. For a US fintech, our team produced roughly 40 delivery automations in about two months and runs an AI QA agent for full-cycle test coverage, all inside an enterprise LLM toolchain with monitoring and guardrails.
- AI products built by Salt Square teams. We are the core development team behind an AI-powered caregiving platform that turns health records into AI summaries and answers questions through a health assistant, and the core engineering partner on an agentic platform for a healthcare AI startup. We also build an AI-powered music platform for hospitality venues that generates playlists from a prompt.
- Agent architecture for digital health. For a US digital health company, our engineers worked on an extensible architecture for AI agents covering provider scheduling, gaps in care and patient outreach.
Where to start
- Decide where AI is worth it, and whether you are ready: AI strategy & readiness
- Move a large legacy codebase faster: AI acceleration tooling build-out
- Ship more with the team you have: AI-assisted engineering
- Put an AI feature in front of users: AI product engineering
- Answer customer questions from your own content: Knowledge & support automation
How we keep AI honest
- Engineers own the output. AI drafts, engineers review. Every conversion, test and automation has a person accountable for it.
- Guardrails from day one. We keep sensitive data out of prompts, keep internal content out of customer-facing answers, and put monitoring on the tools we deploy.
- The method stays with you. We document the rules, prompts and playbooks so your team is not dependent on ours.
Frequently asked questions
Do you build AI models?
We build products and tooling on top of established models, and we connect them to your data and systems. We do not train foundation models.
Which AI tools do you use?
It depends on the job and your policies. Our teams work daily with Claude, Claude Code and Claude Skills, and connect tools to systems through MCP. We fit your approved stack rather than impose ours.
Is our data safe?
We design so it does not have to leave your control: anonymized samples, your own environments, and no customer-sensitive data in prompts unless your policies allow it.