Isaac MineoSenior backend engineer


Senior backend engineer at Accenture's Innovation Garage.

Joined 22 July 2024 as an Analyst. Promoted to Senior Analyst, December 2025.

I build AI products end to end — iOS and web, from the data model to the App Store listing.

70+

proofs of concept with enterprise clients

5

products of my own, two on the App Store

19

repositories, across 169 days with a commit

138

days arbor has traded unattended

Measured 26 August 2026

Open to forward deployed and applied AI engineering roles.

Retrieval over the case studies, the source, and the notes behind them. Answers cite what they read.

Location
New Haven, Connecticut
Focus
AI systems, backend infrastructure
Platforms
iOS and web

Building
Bodivo, RecipeFix, arbor
Latest
Halyard — Docs: the dark suite, the three sweeps, and where dep…
Deployed
3 hours ago

Featured

Designed, built, shipped

KinoLog's Tonight screen showing five film cards with predicted star ratings and the reasoning behind one of them.

01

Live — App Store

KinoLog

A film diary that guesses how much you will like a film before you watch it, and shows you why it thinks so.

KinoLog states a number out loud before you watch a film, and that number is arithmetic, not generation. A content-based nearest-neighbour model runs over your own rated diary — weighting shared directors, writers, cast, subgenre and what the films are actually about — and predicts what you will type, not what the film deserves. Because it is arithmetic it costs nothing, runs in microseconds, and survives the language model being out of credit, which is why the prediction can appear on every card. The language model is handed the finished number and allowed to write the sentence. It is never allowed to produce the value. When you log the film the prediction settles, and the error feeds back as a bias correction — shrunk toward zero and capped at four tenths of a star, so six settled films move it almost not at all.

Platform
iOS, Web
Status
Live — App Store
Demonstrates
Evaluation loops in production
Launched
August 2026
RecipeFix running on iOS, showing the recipe converter with dietary filter chips and a native tab bar.

02

Live — App Store

RecipeFix

Takes a recipe you found online and rewrites it so you can actually eat it, gluten free, dairy free, vegan or otherwise, without ruining the dish.

Every recipe on the internet is written for one diet: the author's. Swapping an ingredient is not a lookup — coconut flour absorbs roughly four times the liquid wheat flour does, so "use coconut flour instead" without changing anything else produces a different and worse dish. RecipeFix adapts the whole recipe, ingredients and quantities and steps and technique notes, and then checks its own output before showing it to you. The checking is the product. A deterministic dietary guardrail blocks anything that plainly violates the diet and triggers one corrective regeneration. A culinary-integrity pass warns about the failures that ruin a dish without being unsafe — an ingredient listed but never used, or a substitution the steps never acknowledge. Scaling is taken away from the model entirely and done in code.

Platform
iOS, Web (PWA)
Status
Live — App Store
Demonstrates
MCP server in production
Launched
April 2026
the order path, and why it is the only one

03

Live — trading

arbor

A program that trades my own money on sports outcomes, works out for itself which bets are worth taking, and keeps an honest score of how it has done.

Kalshi contracts pay a dollar if an event happens and nothing if it does not, so a price is a probability. arbor prices sports events — baseball primarily, tennis at small size — and takes a position where its estimate differs from the market's by enough to be worth the spread. It runs unattended on a schedule, reconciles its fills against its own ledger, and audits itself daily. It is the only thing I have built that is judged by whether it is right rather than by whether it works. Everything else I can decide is finished. This one the market decides, every day, whether or not I agree with it.

Platform
Python, VPS
Status
Live — trading
Demonstrates
Systems accountable to money
Launched
April 2026
worker and scheduled-job architecture

04

In beta

Halyard

Posts about your product to social media on a schedule, without you watching it, and is built so that it can never accidentally post the same thing twice.

Publishing to social platforms on a schedule, unattended, is easy to do badly and dangerous to do wrong. A retry that double-posts to a real account is not a bug you apologise for — the post is live and public, and it is the worst thing this kind of system can do. Halyard is arranged so that it cannot happen. Around that constraint sits the rest of a pipeline: signals about a product become ideas, ideas become content, content is rendered to video with speech and captions, quality gates reject anything that fails, survivors reach a human approval queue, and approved items are scheduled into platform-specific slots and published.

Platform
Web
Status
In beta
Demonstrates
Production engineering
Bodivo's Intelligence screen on a new account, reading "Bodivo has nothing evidence-backed to say about you yet" and "0 of 6 logged" beside what unlocks learning.

05

In development

Bodivo

A training and nutrition app that learns how your body actually responds instead of handing you a fixed plan, and tells you plainly when it does not yet know enough to say.

A training plan is a guess made once, in advance, by someone who has never met you. Bodivo treats the plan as a consequence of a model of the person, revised continuously — and, more unusually, refuses to state anything it cannot evidence. Every metric declares what it is: measured, calculated, or unknown. An unknown is never rendered as zero. A trend below its minimum sample count comes back marked insufficient with a statement saying so, never smoothed into a direction. A view with no evidence returns the reason it is empty and what would fill it, rather than an empty chart. That constraint makes the product emptier early and honest throughout.

Platform
Web
Status
In development
Demonstrates
Evidence-typed state
Sentinel's coverage screen listing ten check areas. Two read Watching, eight read Not watched yet, each with a reason such as no error source connected or no live app URL to render.

06

Report-only, in development

Sentinel

Watches a live app and finds the mistakes that look fine: a contact form that says your message was sent when it never was, a filter that quietly returns everything, a page showing one person another person's data.

Sentinel watches a deployed application and hunts for silent correctness failures: the ones where nothing throws and no alert fires. When it suspects something, it does not report it. A suspicion enters a loop that tries to confirm it against the running system, and a judge, structurally separate from the detector, rules on whether the evidence supports the claim. Only a confirmed finding reaches a person, in a sentence they can act on. Four verbs bound the product: detect silent problems, prevent breakage at the moment of change, verify that things are still right so the absence of an alert means something, and explain all of it in plain language. The surface that best shows the argument is the coverage map, which lists what is being watched and what is not, and gives the specific reason for each gap: no error source connected, no live URL to render, no journeys defined yet. It never reports healthy for something it could not look at.

Platform
Web, GitHub App, Scheduled runs
Status
Report-only, in development
Demonstrates
An agent system built to stay quiet

Client work

70+

proofs of concept contributed to

The Innovation Garage does client-embedded rapid prototyping in two-week sprints. An engagement starts with a business problem somebody can describe but not specify, and ends with working software their own people can use. I lead the client-facing demos and the technical Q&A, which means explaining backend and agentic AI architecture to people who write code and to people who do not, in the same room.

Beyond the prototypes: I built and shipped an internal deployment platform that lets teams deploy behind a security perimeter, on ALB, ECS Fargate, Cognito and AWS Verified Permissions, staging through to production. I am cloud governance lead for the Garage, I built an internal client engagement tool and am running its pilot with real client users, and I made five training videos on agentic AI covering Bedrock agents, AgentCore, knowledge bases and LangChain.

Accenture Innovation Garage

Title
Technology Senior Analyst, Application Engineering
Joined
22 July 2024, as Analyst
Promoted
Senior Analyst, December 2025