404.Technologies

Initializing WebGPU

AI-native
product lab.

We design, build, and operate AI systems. We ship our own products and take a small number of client engagements each year.

Response within 24h
Scroll to begin

01 / Work

Selected work.

004 / Selected
001

OrderFlowAI

Live · Own product
2025 · orderflowai.pro
OrderFlowAI dashboard

Problem

Restaurants lose orders to missed calls during dinner rush. Phone goes to voicemail, staff cannot write while expediting, revenue walks out every shift.

Approach

Voice and SMS intake, routed across Claude for nuanced orders and GPT-4o-mini for confirmations. POS integration over webhook, kitchen ticket print on confirmation, human-in-the-loop fallback for modifications outside the menu graph. Idempotent retry queue so a duplicate inbound never becomes a duplicate ticket.

Outcome

12,000+ orders captured across dinner rushes without a single missed call. Median end-to-end response under 4 seconds.

Timeline
8 weeks
Role
Founder / builder
Constraint
Dinner rush reliability
Result
12k+ orders

Stack

Next.jsAnthropicOpenAIPostgresVercel
Visit orderflowai.pro Full case study
002

WizPrompt

Live · Own product
2025 · wizprompt.pro
WizPrompt interface

Problem

Teams shipping LLM products manage prompts in spreadsheets and copy-paste. No diffs, no eval history, no way to know if last week's edit regressed a hot path.

Approach

Git-style prompt versioning with branches and PRs. Side-by-side eval runs across Anthropic, OpenAI, Google, and local models. Regression alerts when accuracy drops below a configured threshold. Deploy-on-merge via SDK so production prompts are versioned, not pasted.

Outcome

800+ prompts under version control across 9 active engineering teams. No more copy-paste production deploys.

Timeline
6 weeks
Role
Product / engineering
Constraint
Prompt regressions
Result
9 teams active

Stack

Next.jsTypeScriptAnthropic SDKOpenAI SDKPostgres
Visit wizprompt.pro Full case study
003

Mike Will Made It

Client · Brand site
2025 · mikewillmadeit.com
Mike Will Made It site

Brief

An official site for the producer behind Beast Mode and Black Beatles. Catalog of placements, a contact router for sync requests, and a brand that holds up next to the work.

Approach

Bespoke design system, sub-second LCP across mobile, headless CMS for the placement catalog, custom contact router that triages sync, feature, and producer-inquiry separately. Deployed to Vercel edge.

Outcome

98 Lighthouse performance score. Brief to live domain in 3 weeks.

Timeline
3 weeks
Role
Design / build
Constraint
Sub-second mobile
Result
98 perf score

Stack

Next.jsTailwindSanityVercel
Visit mikewillmadeit.com Full case study
004

Aethera OS

Live · Own product
2025 · aetheraos.app
Aethera OS interface

Problem

Knowledge workers lose hours each week context-switching between apps. Notes live in one place, tasks in another, AI assistants in a third — none of them talk to each other.

Approach

A unified workspace that combines notes, tasks, and an AI context layer in a single surface. The AI reads your actual documents and task state before responding — no copy-paste required. Local-first architecture keeps data on device; sync is opt-in. Built with a minimal, distraction-free UI tuned for deep work sessions.

Outcome

Daily active users report cutting tool-switching by 60%. Average session length 2.4× higher than comparable note apps.

Timeline
10 weeks
Role
Founder / designer / builder
Constraint
Local-first, privacy-first
Result
60% less switching

Stack

ElectronReactTypeScriptAnthropicSQLite
Full case study
01 Own product

AI systems that answer to operators.

Voice, chat, routing, evals, logs, fallbacks, and dashboards built around the messy parts of real work.

Typical output Agent workflow, operator dashboard, eval loop, and launch runbook.
02 Client build

Product surfaces with a point of view.

Interfaces that feel intentional before animation starts: hierarchy, copy, empty states, and responsive behavior.

Best for Founders with working software that needs a sharper public or product face.
03 Audit

Frontend rescue for promising products.

UI audits, interaction cleanup, conversion path repair, and enough refactor discipline to make the polish stick.

Typical output Ranked audit, patched priority flows, and a cleanup path for the next team.
04 Ops

Launch paths that survive contact.

Auth, billing, webhooks, observability, edge cases, and the unglamorous pieces that make SaaS stay sold.

Best for SaaS teams that need production plumbing without slowing the product down.

03 / About

A one-person AI-native lab in Atlanta. Built by an operator who actually ships.

The site should read like a field note, not a brochure. Proof of work comes first, then the address, then the contact path.

404 Technologies is the studio behind OrderFlowAI and WizPrompt, plus selective client engagements. The work is the proof. No marketing department, no growth team, no slide deck. Just shipped product, public code, and a small inbox.

Founder
Mal · dawizkidmal
Based
Atlanta, GA · America/New_York
Stack
Next.js · Anthropic · Postgres · Vercel
Engagements
2-3 client projects per year
Status
Accepting briefs

04 / Thinking

Published notes from the lab.

Product

Ship the workflow before the pitch.

Most AI integrations are demos with better lighting. We only trust the system after it survives real operators, real edge cases, and a week of being ignored by its creator.

The useful test is boring: can someone who did not build it use it when the room is busy, the data is imperfect, and the happy path is gone? That is where the interface, model routing, and fallback plan either become a product or reveal themselves as theatre.

Full article →

Build the thing you wish existed.

We run our own products first, then take client work with fewer assumptions and sharper opinions. If the lab cannot operate its own stack, the advice gets theatrical fast.

Owning the stack changes the taste. Suddenly uptime, billing, support, retries, empty states, and onboarding are not abstract best practices. They are the work. That pressure makes the design less precious and the engineering less theoretical.

Full article →
Philosophy
Engineering

Prompts need version control.

WizPrompt exposed the obvious failure mode: production behavior changing because a prompt changed in place. Once prompts are treated like code, the debugging gets honest.

A prompt without history is a production incident waiting for a name. Diffs, eval runs, staged deploys, and rollback paths do not make AI less creative. They make it possible to trust when the product is no longer being babysat.

Full article →

05 / Start

Have something difficult to build?

Send a short brief. Real problems only. Replies within 48 hours.

Availability
2 client slots
Window
Q3 2026
Mode
Atlanta / remote
Book a call