Ryan Yannelli
Portrait,

About

Ryan Yannelli

I have written software since 2009. Most of it was infrastructure: networks, VoIP, data centers, and the operations around them. In 2021 I co-founded Nextvisit AI with Dr. Faisal Rafiq. I am Founding AI Engineer at Lighthouse Legal, with founder Ravi Jayanthi.

Most of my open-source time goes to Laravel and PHP packages, LLM tools, and datasets. I write about where AI helps and where the economics still do not work. I have had Type 1 diabetes since 2023. I run my own fork of Loop, a closed-loop insulin system, on my own body.

Now
Lighthouse Legal
Co-founded
Nextvisit AI
Location
New York, USA
Writing software
17 yrs

My glucose, on my wall clock

The loop

The same 288 samples as the homepage strip, closed into a 24-hour clock. Midnight sits at the top and the hand is now.

12am6am12pm6pm70180

134

mg/dL

—

avg 1454 min ago

A 24-hour clock of interstitial glucose. Current reading 134 milligrams per decilitre, 4 min ago. Midnight is at the top. Radius is glucose; the live edge is the clock hand.

Glucose is the output of that loop. Insulin still acting and carbohydrate still absorbing are the two remaining inputs Loop reports as on board, traced here under the same glucose the clock above is plotting.

4.18U

insulin

3.9g

carbs

carbs · ginsulin · U120 g012 Uovernight
IOB 4.18 UCOB 3.9 gSG 134 mg/dL4 min ago

Loop on-board: 4.18 units of insulin still acting, 3.9 grams of carbohydrate still absorbing, sensor glucose 134 milligrams per decilitre, 4 min ago. The plot under the gauges is twenty-four hours of those remaining quantities. Insulin is below the midline; carbs are above it; the glucose they move is traced underneath.

Background

I started building websites at 12 and registered a business around it in 2009. It was small, most of the work was bad, and I was invoicing for it at 12.

Networks and operations

At 16 I was a network engineer at CSDNET, running infrastructure for hospitals and schools. From 2018 to 2020 I was a professional services engineer at LISS Technologies, doing VoIP rollouts, Exchange migrations, and M365 deployments through their nSpeak subsidiary. In 2020 I moved to PointVoIP as director of operations: data centers, engineering teams, and the infrastructure under a telecom service. I stayed on as a business advisor there through 2025.

Nextvisit AI

In 2021 I teamed up with Dr. Faisal Rafiq, a practicing psychiatrist who had spent most of his career complaining about documentation. We built the thing he wanted: software that writes the clinical note during the visit instead of after it. It handles ambient transcription, note generation, and the billing path behind them for behavioral health practices. In 2025 we went through the LAUNCH Accelerator (Cohort LA35). There is a longer write-up on the Nextvisit page.

Lighthouse Legal

I am Founding AI Engineer at Lighthouse Legal, alongside founder Ravi Jayanthi. AI voice agents for law-firm intake, first for personal injury: they answer the line, screen callers against the firm's criteria, and escalate the ones that qualify. The product is at lighthouselegal.ai, and there is a write-up on the Lighthouse page.

What I work on now

I still do this: get a version into production, then keep the tolerances tight. My stack moved with the work. I started in PHP and JavaScript, added TypeScript and Vue, and now spend most of my time on LLMs and speech models. The open source is mostly Laravel and PHP SDKs, a couple of Vue components, and some datasets on Hugging Face: things I needed once, cleaned up enough to publish. The code is on the work page and the posts are on the writing page.

Timeline

7 milestones

Since November 2023

Type 1 diabetes

>15%

HbA1c at admission

4.9%

HbA1c latest, on the loop

I was diagnosed in November 2023, at 26. I came in at 108 pounds and in diabetic ketoacidosis, spent the admission in the ICU, and had an HbA1c over 15% with a positive GAD65 antibody test.

Since then it has been a control problem with a short feedback loop. I wear an Omnipod 5 pump with Lyumjev and run my own fork of Loop, the open-source closed loop the patient community wrote instead of waiting for a cleared product. My fork adds meal estimates that learn from past outcomes, daily reviews, and workout-aware nights. It reads a continuous glucose sensor and adjusts insulin delivery on a five-minute cycle. The last 24 hours of that feed sit on the clock above (same samples as the homepage strip, plotted on my wall clock). I read it the way I read any production system: trend first, then the outliers, then whether last night's change did what I expected.

Insulin and pump supplies are a cost problem before they are a technology problem, and the software that closes the loop was written by patients. More on that on the advocacy page.

Diagnosed
Loop
My own fork
Pump
Omnipod 5
Insulin
Lyumjev

What I work with

Full résumé

Email

Email is the fastest way to reach me. Founders, engineers, and people running Loop all get answered; cold sales pitches do not.