Work & Case Studies | Devin Pellegrino | Nerority
CASE STUDIES

The Work

Four case studies with the numbers. Regulated medical technology, a live multi-agent operating system, product architecture, and three years of public teaching — each one Problem → Approach → Result.

01 CONSUMER AI PRODUCT · LIVE

Designed the multi-agent anti-hallucination architecture behind 911 PAWS — a live, paid pet-health guidance product.

Role — AI Architect, Pet Portal AI Timeframe — Feb 2025 → in market Domain — Consumer health AI
PROBLEM

A consumer AI giving guidance about someone's pet has zero margin for confident wrong answers — and the person asking is often frightened, at 2 AM, with one question. The system had to be grounded, calibrated, and safe by construction, not by hoping a model behaves.

APPROACH

A 13-agent architecture in which no single model does everything. Specialist gateways convert each raw input — photos, video, audio, vet records — into structured observations before anything is reasoned over; the visual layer is constrained to observation, not inference, with confidence capped by input quality. An event-driven synthesis layer folds every interaction into one cumulative record per pet, so answers start from the animal's actual history — a chart, not a chat. The advisor agent runs a mandatory analyze → formulate → validate pass before it is permitted to answer, and every answer closes with a calibrated confidence and what would raise it. Hard emergency signals are caught by a deterministic check that lives outside the model entirely.

THE PERSON COMES LAST

The priority order is fixed: accuracy, then helpfulness, then personality. The human voice is the final layer, embedded only after the answer has been grounded and validated — so tone can shape delivery but can never override the truth underneath it.

RESULT

Shipped, live, and paid. Designed with veterinarians, selected into the Plug and Play Topeka animal-health accelerator with Hill's Pet Nutrition (2025), and in market at 911paws.com — the one system on this page you can open and use right now.

911paws.com ↗ 13-agent architecture Plug and Play Topeka · Hill's Pet Nutrition
911 PAWS in use: the assistant reasons through Cooper's symptoms, acknowledges that a hands-on exam would tell more, and asks the one question a vet would ask next
911 PAWS in use — grounded reasoning, stated limits, and the question a vet would ask next
13
Specialist agents in the pipeline
4
Input modalities structured before reasoning
1
Cumulative record per pet
4
Reasoning stages before any answer
02 MEDICAL TECHNOLOGY

Trained 50+ clinics and hospitals nationwide on FDA-cleared TMS deployment and plasticity mechanics.

Role — Director of Engineering, Brain Ultimate Timeframe — Oct 2023 → present Domain — FDA-cleared neuromodulation
PROBLEM

A growing medical-device company — FDA 510(k) cleared for MDD and OCD — needed repeatable technical operations: installation, clinician training, documentation, and internal knowledge systems across a complex, regulated product line.

APPROACH

Standardized the install-and-train process end to end, built the internal knowledge and operations architecture on a zero-redundancy standard — every fact lives in exactly one place — and delivered hands-on training on device operation and the underlying neuroplasticity mechanics.

RESULT

50+ clinics and hospitals trained across the country, including Butler Hospital's TMS program — where I personally installed 75% of the TMS wing. A regulated deployment operation that runs as a repeatable system.

FDA 510(k) K230735 ↗ FDA 510(k) K243459 ↗ Butler Hospital · 75% of TMS wing
Devin Pellegrino mid-installation of a Brain Ultimate TMS system at a clinic
On-site TMS system installation
Brain Ultimate team in the field during a deployment
Brain Ultimate field team
03 AI SYSTEMS

A self-referential multi-agent operating system: 37 persistent domain agents, live, under apex orchestration.

Role — Architect & operator Timeframe — 2024 → live today Method — The NERO Method
37
Persistent domain agents, all active
6
Operational planes
242
Versioned session-lineage records
22
Codified propagation laws
PROBLEM

AI agents hold up where a verifier exists — software and mathematics, where the answer compiles or checks. Anything with a verifier can be automated. Outside that, in the open-ended domains that make up most of real work, there is no verifier, so agents drift, forget, and quietly decay.

APPROACH

The NERO Method: a versioned agent lifecycle — instantiate, operate, correct at the exact point of divergence, capture the next version. Progressive complexity ramping. Deterministic rule encoding, where every correction becomes a permanent constraint. Documented publicly across the 22-entry NI-NOTION series.

RESULT

37 persistent domain agents across six operational planes — personal operations, method R&D, community and business ops, and two companies — coordinated by an apex orchestration agent governing by reconciliation, not command. Strict control-plane / data-plane separation, a three-layer state model, and verify-before-write with read-back on every shared surface. Flagship demo: a task agent evolved through 9 versions to autonomously manage 115+ active tasks with zero manual board management.

NI-NOTION methodology series → Reconciliation, not command
04 COMMUNITY & EDUCATION

A founding core of Insiders active for 2.5+ years — set up and operating across their own domains.

Role — Founder, Nerority Timeframe — May 2023 → present Domain — Public education
Nerority Insiders meeting in person
Insiders, in person
500+
Community builders
2.5+
Years of continuous Insider tenure
344★
Advanced-GPTs repository
50+
Recorded sessions
APPROACH

Everything released publicly and dated since May 2023: 33+ GPT agents, 16 prompt libraries, 13 frameworks, a formal curriculum, 50+ recorded sessions, and the NI-NOTION methodology series documenting the NERO Method end to end.

RESULT

A 500+ member community with a founding core of Insiders continuously active for 2.5+ years — each running the method across their own life and work domains. Retention, not reach, is the metric that matters.

Have a Domain That Needs This?

Every engagement is hands-on working time — we build, correct, and ship inside your actual systems. You leave with a working system and the ability to operate it.