About
Isaac Mineo — Backend & AI engineer
- Role
- Backend & AI engineer
- Location
- New Haven, Connecticut
- Middlebury
- 2019 to 2023
The short version
I build backend and AI applications at Accenture's Innovation Garage, and independent products outside work. My projects start with everyday questions: what should I watch, what can I cook, and why did my keyboard change that word?
Accenture
Backend and AI engineering
Middlebury
Neuroscience
AWS
Enterprise application development
iOS + Web
Independent products
For a conversation about my work, get in touch.
From neuroscience to software
I studied neuroscience at Middlebury College and graduated in May 2023. My software projects now put particular emphasis on evaluation: what a system predicted, what actually happened, and what should change next.
That interest shows up in my projects. KinoLog records a prediction before a film is watched. Idio distinguishes touch evidence from a guess about language. Bodiivo separates a proposed change from permission to apply it.
What I build at work
I joined Accenture in July 2024 and was promoted to Senior Analyst in November 2025. My official title is Technology Senior Analyst, Application Engineering. I work on backend and AI applications in the Innovation Garage.
AWS is part of my daily development work: Bedrock and AgentCore, S3 and DynamoDB, container deployment with ECR and ECS, and access controls with Cognito, Verified Permissions and KMS. I also work with Python, FastAPI, retrieval systems and agent integrations.
My responsibilities include turning an uncertain problem into a working proof of concept, making the backend decisions, pairing with other engineers and explaining the result. Public examples here omit client names and confidential engagement details.
The products are personal
RecipeFix and KinoLog are available on the web and App Store. Idio, Smart Rant, Whereabouts and Bodiivo explore different problems in native interfaces, personal data and adaptive software. Arbor v2 is private research work, not a public consumer release.
I care about the parts after the demo too: a useful first screen, a recoverable failure, a clear explanation, and a way to learn from what actually happened. Each project page separates implemented work from what still needs verification.
How I decide
Opinions, each earned somewhere
The important number should come from the system, not the explanation.
KinoLog computes a rating locally and lets a model help explain the recommendation. The explanation is not allowed to invent the star value.
A suggestion and permission to act are different things.
Bodiivo puts proposals through quality checks, arbitration and an explicit autonomy policy before a writer applies a change.
Missing evidence should stay missing.
The project dossiers state their limits. Ask uses an explicit source set, and unavailable facts should not turn into plausible answers.
Build the useful thing. Then build the check that tells you whether it worked.