Some parts of this page need JavaScript.

Skip to main content

Analytics only if you accept. Privacy

AI infrastructure, software and public code

The technology we build with

Below are the platforms, AI systems, cloud patterns and public repositories behind the work.

  • Public source code

    JarvisNano, Verrow and Kintunnel have public source linked from their case studies. Other work is documented through its architecture, implementation and available product demonstrations.

    Read the code
  • Case studies

    Write-ups of what each system does, how it is built and what it changed.

    View the work
JarvisMCP, running in production
In production · JarvisMCP
How we built it

In short

AI infrastructure and model tooling from a Fort Lauderdale studio: builds on Claude, OpenAI and Gemini models and self-hosted open-weight models, connected with retrieval and MCP servers, and deployed behind server-side token brokers. Some of the source is public on GitHub.

How it runs

How a request moves through what we run

The same layout runs our own products: a router in front, services in containers behind it, and the data stores underneath.
01 Client02 Router03 Services04 DataBrowserFlutter appAPI clientTraefik on DockerReact or Next.jsNode APIPhoenix servicePostgreSQLRedisS3 storage

Stack

What we build production systems with

AI models

LLM workflows, retrieval, evaluation, tool use, and private-model planning when the project needs more control.

Claude / OpenAI / Gemini / open-model evaluation / model routing / RAG / evals / function calling

Agent infrastructure

Human-supervised agent systems that can research, code, operate tools, and keep context across real work sessions.

Claude Code / MCP servers / JarvisNano / browser automation / GitHub tools / approval gates

Applications

Production web and mobile software with performance budgets, load testing where it matters, and agreed deadlines.

React 18.3 / TypeScript 5.9 / Vite 8 / Vite+ / Next.js / Node / FastAPI / Phoenix / Flutter

Data and automation

Operational systems that connect CRMs, inboxes, forms, payments, documents, and internal dashboards.

PostgreSQL / Supabase / Redis / GoHighLevel / HubSpot / Stripe / n8n / Make / Zapier / webhooks

Cloud and ownership

Deployments designed around portability, observability, security, and the right amount of managed service.

Docker / Traefik / GitHub Actions / Dokploy / private container registry / Vercel / AWS / S3/MinIO / hybrid cloud

Interfaces and hardware

Experimental systems where agents meet devices, voice, video, robotics, and real-world inputs.

agent control loops / embedded AI prototypes / ESP32 / voice agents / video understanding / WebRTC / WebSockets

How this helps clients

Software you can run after handoff

We explain architecture decisions as we make them, and hand over documented code your team can still read six months after launch.
  • Start with the business workflow, then choose the stack.
  • Use managed platforms when speed matters and portable systems when control matters.
  • Keep AI systems observable, reviewable, and grounded in real company data.
  • Document architecture and tradeoffs so clients can maintain the code.

FAQ

Frequently asked questions

Common questions about our AI infrastructure, models, MCP servers, and self-hosting.

What AI models and providers does Ingenious Digital work with?

Claude, OpenAI and Google Gemini models (including the Gemini Live API behind our JARVIS voice assistant), and open-weight models we can host ourselves. We choose the model by task, cost and speed, and connect it to your data through retrieval (RAG) and tools (MCP) so it answers from your records.

Do you build MCP servers?

Yes. Model Context Protocol (MCP) servers give AI assistants controlled access to tools, APIs and internal data, so an agent can look up a record, start a workflow or filter a view. Our JarvisNano prototype calls tools through our JarvisMCP gateway.

Can you self-host models instead of using a third-party API?

Yes. When data residency, cost at volume or speed call for it, we run open-weight models on your cloud or your own servers. When a hosted model is the better choice, we call it through a server-side broker so keys never reach the browser.

What is RAG and do you implement it?

Retrieval-augmented generation (RAG) makes an AI answer from your own documents and data, which reduces made-up answers and keeps them current. We build the whole pipeline (ingestion, chunking, embeddings, vector search and re-ranking) and connect it to assistants, search and internal tools.

How do you keep AI features secure and reliable in production?

Keys stay on the server behind short-lived token brokers, model calls run behind rate limits and test suites, and agent actions are limited to the MCP tools we allow. We test model behavior before release; our public DRQ Benchmark project shows how.

Work

Related case studies

JarvisNano, a desk AI hardware prototype; the harness our AI agents ship code through; and Asteria, a birth-chart guide.
  • JarvisNano, running
    AI & Machine Learning

    JarvisNano

    An open-source embedded AI project with C/C++ firmware on the Waveshare ESP32-S3 Touch AMOLED 1.75C and hardware-first prototyping to explore assistants that run on a device. A Kotlin Android app is the optional companion.

    Read the case study
  • Developer Tools

    Agent delivery harness

    Every change on this page went through the same pipeline: a board item becomes an agent working in its own isolated copy of the code, then a deterministic gate, then a review by a model from a different vendor, and only then does it land.

    Read the case study
  • Asteria, running
    Consumer Productivity

    Asteria

    Asteria explains a person's birth chart in plain language and keeps calculated fact visibly separate from interpretation. A Rust calculation engine does the astronomy; a React client makes it readable for complete beginners.

    Read the case study

How we work

How a change reaches production

The same five gates for our own products and for yours. Nothing reaches your customers on the strength of a demo.

  1. 01 · Isolate

    Its own branch and workspace

    Every change starts in a separate copy of the code, so work in progress never touches what is live.

    git worktree add .worktrees/your-change
  2. 02 · Build

    Engineers with AI agents

    Agents write and test code beside us. They hold no passwords: every call they make goes through our gateway, which checks it and logs it.

    agent → gateway → policy · audit · call
  3. 03 · Prove

    Evidence, then review

    Tests, before-and-after screenshots of the running app, then independent reviewers. It does not move on until no serious finding is left.

    review: 0 high · 0 medium → pass
  4. 04 · Ship

    Deployed and checked live

    The pipeline builds and deploys. Then we check the running service itself, not just a green build.

    replicas 1/1 · image sha matches · health 200
  5. 05 · Watch

    Checked every night

    Each integration we run is tested against its provider every night, so a changed API is caught before your team notices.

    nightly: each integration vs its provider

Talk to us about your stack

Tell us what you run today and what you need it to do, and we will suggest where to start.