pro
Kevin Minn
Full stack engineer. Healthcare AI, FHIR,iOS, and Security.
- 1
- Followers
Project AIR is an open-source toolkit that keeps a tamper-proof record of what AI agents do. Every action an agent takes gets signed, chained to the step before it, and tied to the person who authorized it. The chain is anchored to Sigstore Rekor, a public transparency log, so anyone can check later that nothing was changed or deleted. When something goes wrong, you can show exactly what happened and who approved it, instead of just saying it happened.
It also runs 16 detectors over every record, covering all ten risks on the OWASP Top 10 for Agentic Applications, three from the OWASP list for LLM apps, and three of its own. It can stop an agent mid-action until a person signs off. It plugs into LangChain, LlamaIndex, and the OpenAI, Anthropic and Gemini SDKs with a line or two of code.
I built the Python library and command-line tool, published on PyPI as projectair (now at version 1.3.1) under the Apache 2.0 license, plus the website and dashboard in SvelteKit. We run it on our own infrastructure too: every request to vindicara.io (http://vindicara.io) is recorded and anchored to Rekor every minute.
5
2
64
Motium is an AI interviewer for hiring. Candidates sit down face to face with Wren, a talking video interviewer, and have a real conversation about the role instead of filling out a form. Wren is paid by the employer but works for the candidate: the job is to find what each person is unusually good at and make sure the hiring manager sees it. Every role interviews for the same fixed skills, so candidates can be compared fairly, but the questions are written fresh for each person.
After the call, the hiring manager gets a clear recommendation with the reasons behind it, plus a video recording of the interview. Every question, answer and verdict is signed and hash-chained, so the record can be checked later and can’t be quietly changed. Fairness rules live in the code, not in a policy doc: nothing is scored before consent, and an interview can’t be cut short early without a review.
I built the server that runs Wren’s side of every turn, the live video face running on GPUs, the employer dashboard, Studio 11 (a short spotlight show for recognizing employees), and Pete, a built-in help assistant. It runs on Claude through the Claude API and Amazon Bedrock, with ElevenLabs for voice.
0
35
Nourished by Mira is an iOS nutrition coach for people who want to lose fat without losing muscle, especially people on GLP-1 medications like Ozempic. It builds protein-first meal plans and grocery lists, tracks body composition, and adjusts portions around where someone is in their medication cycle, so the plan still works on days when appetite is low.
Mira, the in-app coach, writes the meal plans and talks people through their day, with a real voice. The app also tracks medication doses with reminders, can scan a prescription bottle label, and can put together a report to share with a doctor.
I built the whole app in Swift and SwiftUI, plus the backend on AWS: serverless functions, a user database, sign-in, and the AI that writes meal plans on Amazon Bedrock. It’s live on the App Store.
0
30
Amyneion helps community doctors bring autoimmune clinical trials to their own patients. It pulls recruiting trials from ClinicalTrials.gov (http://ClinicalTrials.gov) and the EU trials registry and turns each trial’s eligibility rules into checks that a person reviews and approves. Then it reads a practice’s health records over FHIR and shows the care team which patients qualify, or are one lab or one visit away. Every yes, no or unknown comes with a plain-language reason, and missing data gets flagged instead of quietly ruling someone out.
I built all of it: the matching engine, the trial ingest, the FHIR reader, a navigator case book, the clinic Console and the sponsor reporting. It runs on Python, FastAPI and PostgreSQL, hosted on AWS. It can also run fully offline on a clinic’s own machine, with notes read by a model on that machine’s hardware. It has been tested against the Epic and Oracle Health sandboxes, and everything runs on synthetic patients. No real patient data has ever touched it.
0
49