Available · Q3 2026 · Remote / Hybrid

Elwood
Berry

Agentic Automation Engineer
& AI Solutions Architect

I design and ship agentic workflow systems — n8n orchestration, schema-validated LLM pipelines, and full-stack Next.js delivery. Every capability on this page maps to a public demo build you can open and inspect, not a claim you have to take on faith.

Open to Opportunities
19
Demo Builds in the Portfolio
6
Industry Verticals Covered
2
Builds in Public Preview
0
Fabricated Metrics on This Page
Core Stack
n8nOpenAI APIAnthropic APINext.js 14TypeScript (strict)VercelNode.jsREST / WebhooksWix Studio + VeloAWS S3 / LambdaHubSpotSlack API

From Intake to Audit Trail

Every one of the 19 portfolio builds follows the same discipline: unstructured input in, orchestrated processing in the middle, a traceable decision out. Some builds carry an AI layer. Seven deliberately don't — because the honest answer to some workflows is a deterministic pipeline, not a model.

Workflow & Process Automation
n8n orchestration, webhook routing, event-driven pipelines, CRM sync
Agentic AI Systems
LLM classification, schema-validated extraction, confidence scoring, RAG-grounded lookups
Deterministic Pipelines
Knowing when a workflow needs no AI at all — and labeling it that way
Full-Stack Delivery
Next.js 14 static export, strict TypeScript, Vercel, config-driven content
01 — Intake & Triage
Unstructured In, Routed Out
Claims, contracts, invoices, leads, tickets — documents arrive messy, get parsed, extracted against a fixed schema, and routed to the right downstream queue with a confidence score attached to every decision.
02 — Orchestration
Pipelines That Handle Their Own Failure
n8n workflows connecting webhooks, LLM calls, storage, CRMs, and messaging — with confidence thresholds, human-in-the-loop approval gates, and reflection loops that catch bad outputs before a customer does.
03 — Traceability
Every Decision Writes an Audit Record
When a document lands in a queue, the trail shows exactly which extracted evidence sent it there. Timestamped audit logs are part of the architecture in every build, not a compliance afterthought.
04 — Honest Engineering
Unknowns Render Visibly
Statuses, TODOs, and unfinished decisions render on-page instead of shipping silently or crashing. If a build isn't live, its page says so. That discipline is enforced in the type system, not the marketing.

Nineteen Builds.
Six Industries.

Browse All 19 Builds

A selection below. Each build ships with its own page: architecture diagram, layer breakdown, stack, and a representative payload — labeled mock where it's mock.

Legal & Compliance ServicesPreview

Document Ingestion & Analysis

Discovery bundles and lease packages arrive as single massive PDFs. The pipeline splits them, extracts key entities against a fixed schema, and routes each document to its correct queue — with the sorting decision always traceable back to the extracted evidence.

n8n · OpenAI API · Next.js · Vercel
Mid-Market SaaS & Tech StartupsPlanned

Self-Correcting Support Agents

Single-shot LLM support agents fail silently. This design deploys autonomous agents with built-in reflection loops that detect failed responses and re-route automatically — every response, reflection pass, and re-route logged, so self-correcting is provable, not a slogan.

n8n (AI nodes) · OpenAI API · Next.js · Vercel
Banking, Finance, FinTech & InsurancePlanned

Fraud & Risk Monitoring

Ingests live transaction streams via webhook, cross-references historical pattern vectors, scores each anomaly, and fires instant Slack and email alerts with the pattern evidence attached — so risk teams triage signal, not noise. Deterministic pipeline: no AI layer, by design.

n8n · Next.js · Vercel · Slack · No AI layer — by design
Automotive Retail & Dealership OperationsPlanned

VoltFlow Delivery Assistant

Replaces the hand-repeated 45-minute EV delivery walkthrough. Cross-references the buyer's zip against regional EV parameters, auto-generates a custom charging map, and files rebate links electronically at handoff — consistent for every buyer, logged for every delivery.

Next.js · Vercel · n8n · No AI layer — by design
How We Work Together

Built for
Enterprise
Scale

Whether you're placing me with a client full-time or need targeted automation architecture for a defined scope, I integrate cleanly with existing teams, governance structures, and audit requirements.

Start the Conversation
Full-Time / Contract Placement
Preferred
Embedded engineer on your team or your client's team. Target roles: Applied AI Engineer, Agentic AI Engineer, Automation Engineer, AI Solutions Architect, Senior Full-Stack Engineer. W-2 or 1099.
Project-Based Engagement
Available
Scoped delivery through Adbyrd, LLC: workflow architecture, agentic pipeline builds, or production migration. Fixed milestones, defined handoff, documentation included.
Fractional Automation Leadership
Advisory
Part-time strategic leadership for teams building automation capability without a full-time hire — architecture reviews, vendor evaluation, and workflow governance.
Location & Travel
Remote-first (US). Based in Dallas–Fort Worth, available for onsite in DFW and flexible for client travel.
Demonstrate, never claim. Every capability on this page maps to a build you can open, inspect, and question. No invented metrics. No imaginary clients. No testimonial in this slot — because I won't fabricate one to fill a wireframe.
The operating standard behind the portfolio
01
Demonstrate, Never Claim
Numbers on this site are auditable against the portfolio's master config. Mock data is labeled mock. Client names stay out of copy until deals close.
02
Unknowns Render Visibly
Unresolved decisions ship as explicit, on-page TODOs — never silently omitted, never allowed to crash a page. Enforced at the component level.
03
The Right Tool Is Sometimes No AI
Seven of the nineteen builds are labeled 'deterministic pipeline — no AI layer' as a deliberate, auditable choice. Knowing when not to use a model is the job.
04
Production Discipline From Line One
Strict TypeScript, build-time type safety, config-driven content, defensive component patterns. The portfolio's own codebase is the first case study.
Let's
Talk.

Actively responding to opportunities for Q3 and Q4 2026 placement. Recruiters and hiring managers: I'm happy to walk through any build's architecture live, provide references, and prepare a capability brief tailored to your client.