Sheet 1 of 7Idea · Concept · Prototype · ShippedApplied AI · Full stack
Applied AI and full stack engineering

From idea to concept
to shipped product.

We take your idea or your problem to a secure, tested product in production: AI voice agents, AI features in your product, and web and iOS apps people enjoy using.

Book a callSee how work moves

From the engineer who built Receptiva, an AI phone receptionist live since June 2026, after 13+ years building enterprise software.

Receptiva call flow shown four ways: a rough idea, a precise concept, a wireframe prototype, and the shipped product with a live call.caller, EN or ES?who answers at 2am?hot case: get an atty!lead lands in their apptest every change102112104106108110INCOMINGCallerES · ENReceptiva agentLIVECALLERMe chocaron esta mañana.AGENTLo siento. ¿Está herido?NEW CASEURGENTROUTING TO ATTORNEYWarm transferAttorney on the lineSIP connectedNew leadAuto accident · ESIntake capturedIn the firm dashboardEVALS
FIG. 1 Receptiva call flow, idea to productionDrawing
13+ yearsbuilding enterprise software at scale
June 2026Receptiva live in production, answering real calls
2consumer apps of our own on the App Store
70U.S. patents issued to the founder
Sheet 2 of 7FIG. 2 to FIG. 5 · One product, four stagesProcess
How work moves

Every product starts as an idea.

AI lets us explore more concepts in a day than we used to in a month. Engineering judgment decides which one gets built. You see each stage and sign off before the next one starts.

  1. FIG. 2 · 01 Idea

    Frame the problem

    • Who it serves, and what has to be true for them
    • Constraints, risks, and what success looks like
    • The smallest version worth shipping

    Out: a one page problem brief

  2. FIG. 3 · 02 Concept

    Explore, then commit

    • Several directions explored quickly with AI
    • Architecture, data model, and threat model
    • An eval plan for anything a model decides

    Out: a chosen concept and a fixed scope

  3. FIG. 4 · 03 Prototype

    Make it real enough to test

    • Clickable flows on real data
    • Real model calls, scored by the eval suite
    • Your team tries it before we harden it

    Out: a working prototype and eval results

  4. FIG. 5 · 04 Shipped

    Run it in production

    • CI/CD with Terraform plans on every pull request
    • Monitoring, alerting, and an on call runbook
    • Access controls and audit logs from day one

    Out: software your users rely on

An idea is cheap. Software that stays secure, tested, and running for real users is the hard part. That part is our job.

Sheet 3 of 7FIG. 6 to FIG. 10 · What we buildCapabilities
What we build

What we build, and keep running.

Each capability is backed by software we run today. Blue marks the parts that are live.

FIG. 6 · AI voice and chat agents

Agents that answer the phone and hand off to a person.

Inbound intake, triage, scheduling, and follow up, in English and Spanish. We pick the voice stack per use case and test call behavior before every deploy.

Proof
Receptiva, answering law firm calls since June 2026
Stack
Telephony, SIP transfer, and realtime voice models, chosen per use case
Gate
An eval service scores call behavior before each release
FIG. 6 Voice agent with a swappable model layer and an eval gate
FIG. 7 · Secure by design

Defense in depth, built in from the start.

Security starts at the foundation, not the login screen. Layers of protection surround what matters, each one narrows the attack surface, and each assumes the one outside it can fail, so a single mistake never exposes a record. We threat model before we build and keep watching after we ship.

Principles
Smallest attack surface, least privilege, deny by default
Stay ahead
Threat models up front, scanning on every change, alerts on unusual access
Proof
Medical case management platform: closed privilege escalation and unauthenticated access paths
Now
In progress SOC 2 and HIPAA readiness on two production platforms
FIG. 7A Plan view: rings of protection around what matters
FIG. 8 · Web and iOS apps

Consumer apps people come back to.

Native iOS and web apps, designed screen by screen and shipped through App Store review. We build and run two of our own, so we know the whole lifecycle, from first idea to updates in production.

Proof
BookMatcher: AI book recommendations on iOS and the web
Proof
Recall Beacon: product safety recall alerts with push notifications
Stack
SwiftUI, Next.js, Hono, Supabase, Firebase, RevenueCat
streak?bestsellers01 to 04IDEASHIPPED · APP STORE
FIG. 8 BookMatcher Discover: the first idea, and the real screen on the App Store
FIG. 9 · AI features in your product

Put your product inside Claude and ChatGPT.

LLM features, MCP servers, and evals for the product you already have. Your customers ask questions in their assistant, and your system answers with its own data and permissions.

Proof
Receptiva's public REST API and MCP server, used from Claude and ChatGPT
Auth
OAuth 2.1 with PKCE, scoped per firm
Platform
Claude Platform, OpenAI, and Gemini
FIG. 9 An assistant calls your product through an MCP server
FIG. 10 · AI native delivery

We build the machine that builds the software.

Every task starts as a spec: what to build, the tests that define done, and what not to touch. Agents work from it in parallel, and automated gates check every change. Small changes that are easy to roll back ship on their own. Anything that touches security, data, money, or infrastructure goes to an engineer, who owns the specs and spends judgment where it matters.

Specs
Written intent and tests first, so agents build the right thing
Gates
Tests, evals, security scans, and a risk score on every change
Humans
Engineer review for high risk changes and anything new
Learn
Every incident and review tightens the rules the agents follow
FIG. 10 Spec driven, risk routed: gates decide what needs a person
Sheet 4 of 7Selected work · status as of November 2026Work
Selected work

Shipped and running.

Client platforms, from law firms to a local septic company, and our own products. Each one is stamped with where it stands today.

Legal · ClientLive since June 2026

Receptiva: bilingual AI phone receptionist for law firms

Answers every inbound call in English or Spanish, triages the case, and warm transfers urgent ones to an attorney. Eval suite, multi tenant isolation, MCP server.

Read case study
Healthcare · ClientSOC 2 and HIPAA in progress

Security and delivery overhaul of a medical case management platform

Unblocked launch, closed privilege escalation and unauthenticated access paths, and moved every deploy to a reviewed Terraform plan on AWS.

Read case study
Home services · ClientIn daily use

AOTL Septic Service: field and back office operations app

Replaced paper and clipboards for a multi truck septic company. Techs capture jobs on site, customers get same day service reports, and critical issues reach the office in Slack within seconds.

Read case study
Consumer · Our productOn the App Store

BookMatcher: AI book recommendations on iOS and the web

Personalized shelves and recommendations from your reading history, built and run end to end by CodeHarmonic.

Read case study
Consumer · Our productOn the App Store

Recall Beacon: product safety recall alerts

Push alerts the moment a food, vehicle, or medicine recall matches what you follow, with family sharing and in app purchases.

Read case study
Sheet 5 of 7Parts list · the toolboxParts
Parts list

The parts we build with.

Chosen per project, not by habit. Each part here runs in production today.

RefPartWhat it does for you
AI and agents
201Claude Platform and MCPYour product as a tool inside Claude and ChatGPT, with scoped access
202Realtime voice and telephonyPhone agents with warm transfer to a person
203Eval suitesModel and prompt changes scored before they ship
Cloud and data
211AWS, Google Cloud, TerraformInfrastructure as reviewed code, previewed on every pull request
212Postgres and SupabaseDeny all row level security and tenant isolation tests
213Durable jobs and queuesBackground work that retries instead of getting lost
Delivery and operations
221CI/CD pipelinesAutomated tests, plans, and deploys on every merge
222Monitoring and alertingErrors become tickets, alerts reach a person
223AI coding agentsTest first pull requests, directed and reviewed by an engineer
Apps
231SwiftUINative iOS apps through App Store review
232Next.js, TypeScript, PythonWeb apps, dashboards, APIs, and agents
233Stripe and RevenueCatSubscriptions and in app purchases
Sheet 6 of 7How we work together · Who you work withAbout
How we work together

Start wherever you are.

Bring an idea, a problem, or something that is already broken. We work in whatever shape fits your team.

01 · Build

Take it from idea to shipped

A defined outcome, scope, and timeline, agreed before work starts. Best for a first product, a voice agent, or an AI feature.

02 · Fix

Rescue what is stuck

Broken deploys, security gaps, or a codebase nobody wants to touch. We find the root cause, fix it, and harden it.

03 · Advise

Get a clear recommendation

Architecture reviews, AI strategy, and build or buy calls, written up so your team can act on them.

04 · Embed

A senior engineer on your team

On call for your team, part time or for a stretch of work. You own the code and the infrastructure.

Founder · Principal engineer

Gabe Fernandez

Experience
13+ years in enterprise software
Degree
M.S. Computer Science, deep learning
Patents
70 U.S. patents
Works
Remotely, across the U.S.
About

Engineering discipline, from first idea to production.

Gabe has spent 13+ years building enterprise software, including platforms that handle over a million requests a day. He holds an M.S. in Computer Science focused on deep learning, cofounded and built Receptiva as CTO, and leads engineering for a medical case management platform. See his full background.

  • Systems thinkingSee the whole system before changing one part.
  • ReliabilityTested changes, safe deploys, and fast recovery.
  • AvailabilityMonitoring and on call that reach a person.
  • SecurityLeast access by default, audited when it matters.
  • ScaleDesigned for the next order of magnitude.
  • Clean codeCode the next engineer can read and change.
Sheet 7 of 7Start a projectContact

Bring the idea, the problem, or the challenge.

We will help you shape it, solve it, and ship it. Tell us what you are working on, and we reply within 24 hours.

Project inquiry* required