AI can write the code.
That doesn't mean the system is engineered.
I'm David. I've spent more than 24 years building and debugging systems that had to work. Now I teach people to engineerthe software they're generating, help businesses put AI and custom LLMs to real use, and fix or build applicationswhen another prompt isn't enough.
24+ yrs
Engineering, security, R&D
CTO-level
Technical leadership
Direct
You work with me
"The language is the paint. The fundamentals are the brush." Cybersecurity, distributed systems, R&D, technical leadership. I'm a nerd, but more importantly, I'm a debugger.
David, FounderRead the full brief →
Three ways to put real engineering on your problem.
Pick the one that matches where you are. Not sure? That's what the discovery call is for.
Learn to Engineer with AI
ForBuilders who can generate an app but can't yet explain why it works, or why it doesn't.
Not a course. We work on your actual application and I teach you how to engineer it. Let AI write the code if you like. You still have to understand what you're building and how the pieces fit together.
- Architecture and system design, on your real codebase
- Databases, APIs, authentication, and infrastructure
- Security, testing, debugging, and deployment
- Driving Claude, Codex, and other coding agents, not the other way round
- Reading and validating what AI-generated code actually does
- Avoiding the decisions that end in expensive rewrites
Business AI Consulting & Custom LLM Solutions
ForCompanies that want AI to do real work, not a demo for the board.
I find the problems where AI actually pays off, design the system, and build it: custom LLM applications, private or locally hosted models, retrieval over your own knowledge, agents wired into the business you already run. Part of the job is telling you when not to use AI.
- Feasibility, architecture, and implementation strategy
- Custom LLM apps, RAG over business knowledge, agents, integrations
- Private and locally hosted AI systems
- Model, hardware, infrastructure, and operating-cost evaluation
- Proof-of-concept through production
- Technical advisory and fractional CTO engagements
$99 to start
One-hour discovery call, credited toward the engagement. Scoped and priced before any work begins.
Custom Engineering, Architecture & Code Rescue
ForAnyone holding a broken, fragile, or half-finished application, or a real idea that needs building correctly.
Diagnose and rescue what's broken. Redesign what was never architected. Take a promising prototype into production, or build the whole thing from scratch. Performance, integration, and infrastructure problems included. If it's experimental, even better.
- Diagnose and rescue broken or fragile applications
- Review and redesign poor architecture
- Complete custom builds, prototype to production
- Performance, integration, and infrastructure fixes
- Lead complex technical implementations
- Experimental systems and applied R&D
From $500
Rescue from $500 · Builds from $2,500. Fixed scope and price, agreed before I start.
You work directly with me.
Every engagement comes to me. I review it, scope it, and in most cases do the work myself. No roster, no account manager, no handoff to someone you've never met.
David
Founder & engineer, SoloNerds
"The language is the paint. The fundamentals are the brush."
My career has run across cybersecurity, distributed systems, systems engineering, R&D, and technical leadership, including CTO and technical lead roles and years as Officer in Charge of U.S. Navy Cyber Protection Teams. The common thread: systems that had to work, and getting pulled in when they didn't. Web apps, sensors, automation, neural nets, LLM pipelines. The "impossible" problems were usually the ones that landed on my desk.
Today I split my time between client work and advanced local AI systems and independent engineering research. That's where the AI advice comes from: running the models, not reading about them.
I don't just hand back a fix.When we work together I explain what I'm doing and why, so you're not dependent on me, or on the next prompt.
System design
How the pieces fit, not just making one work in isolation.
Debugging instinct
Years of reading logs, tracing state, building the real mental model.
Honest trade-offs
When to optimize, when to ship, which corners are safe to cut.
Safe and scalable
Works today, and still works when ten times the users show up.
Want to see how I think before you book? My personal research site has the field notes and the R&D, from JARVIS Oracle to agent blueprints and security tooling. zerobandwidth.com
The kind of systems I actually build.
Not a résumé. A sample of what's on my bench, labeled honestly: shipped production work, applied engineering, and experimental research that's still in progress.
SoloNerds platform
This site. Next.js and MongoDB, AWS Cognito auth, Stripe and Stripe Connect payments, a zero-knowledge encrypted vault for credentials, a project room with milestones and PR-based delivery, and a Zoom post-meeting pipeline. Designed, built, deployed, and operated by me.
Custom software platforms
Full-stack platforms for businesses: authentication, payments, workflows, third-party integrations, infrastructure, and deployment. The boring parts done right, which is what makes the interesting parts possible.
JARVIS Oracle
A private, locally hosted AI system I use as a test bed for models, retrieval, and agents, and for the hardware, latency, and cost realities of running AI without a cloud API. The lab behind the advice I give on business AI.
Local AI film generation
Generative video pipelines on local hardware: model orchestration, scheduling, and the unglamorous engineering that keeps long-running GPU jobs reliable and reproducible.
Computer vision
Vision models and pipelines for detection and analysis, from training data through inference on real hardware, where the camera is never as clean as the dataset.
Forensic analysis
Tooling and process for reconstructing what actually happened in a system from logs, state, and evidence. Honed leading U.S. Navy Cyber Protection Teams, and the same instinct I use on a broken codebase: find the real cause, not the convenient one.
PRODUCTION is shipped and running. APPLIED is engineering I do on real problems. R&Dis experimental and ongoing. I won't blur the three.
The write-ups behind this work live on my personal site, zerobandwidth.com.
How I think, in writing.
Long-form pieces on engineering, AI systems, security, and leadership. Published on LinkedIn and my research site, written the way I talk on a call.
Before You Build the AI Agent
An eight-part series on the groundwork most teams skip, from source-of-truth maps to logging.
Start hereYou Don’t Build AI Agents by Throwing Documents Into a Chatbot- 01Build the source-of-truth map
- 02Turn tribal knowledge into structured knowledge
- 03Create decision trees and escalation logic
- 04Build the knowledge base correctly
- 05Decide what should be RAG, fine-tuned, or rules-based
- 06Build an evaluation harness before production
- 07Keep humans in the loop where it matters
- 08Log everything
Field notes from the lab
Longer, rougher, and closer to the research.
- Trustworthy AI Agent Blueprint, Part I: Bounded Autonomy
- The ZeroBandwidth Principle
- The Birth of Neon Steelwave
From first conversation to working software.
No 47-step onboarding. No bidding. No meetings about meetings.
Start the conversation
Book a $99 discovery call or send a message. Tell me what you're building, what's broken, or what you're trying to decide. I read it myself, and I'll tell you honestly if I'm not the right fit.
I assess and recommend
Instruction, an architecture or AI review, or a scoped engineering engagement. One of those will fit. Sometimes the answer is "you don't need me yet," and I'll say so.
We agree on scope and price
Fixed scope and price before any work starts. Instruction is booked per session. Projects run on milestones in a project room with updates and PRs you can actually read.
We build, and you understand it
I explain the what and the why as we go. Whether I'm teaching, advising, or doing the work, you end up with a system you can reason about, not a black box with my name on it.
Not ready to talk yet? Generate a free project blueprint and bring it to the call. It's a planning tool, not a sales funnel.
$99 Discovery Call
One hour with me. Bring the codebase, the idea, or the AI question. I'll assess where you are, tell you what I'd do, and give you a clear next step, whether that's instruction, a scoped project, or nothing at all.
- 60 minutes, video call, with me directly
- Honest assessment, no upsell script
- Fee credited toward any engagement you book
Hire the engineer, not the prompt.
Whether you want to learn to build correctly, put AI to work in your business, or get a difficult system fixed or built, it starts with one conversation. With me.
A personal reply within 24 hours
I read every message. There's no bot.
Fixed scope before work starts
No scope creep, no surprises
No surprise invoices
Prices agreed up front, always
No spam, ever. Your code and your data stay private.