JKP Software · Joseph Pereira

AI development · Spokane Valley, WA · Remote

Software at AI speed.
Engineering you can actually trust.

I'm Joseph — a freelance developer who ships LLM features, chatbots, and full products in weeks using AI tooling, backed by real production engineering: commercial products I build, sell, and operate myself. The AI writes fast. I make sure it's right, secure, and still running a year from now.

Start a project See case studies

Fixed scope · Fixed price · You own the code from day one

Certified
Foundational C# with Microsoft
Since 2021
Selling & supporting my own products
In production
LLM chatbot with tiered model routing
End to end
Design → build → deploy → operate

Services

Engineering solutions for modern businesses

Before any proposal or financial commitment, we’ll schedule a discovery session to understand your goals, evaluate the project’s requirements, discuss the technical direction, and confirm we’re aligned on scope, expectations, and execution.

AI features for your product

Customer-facing chatbots, document processing, semantic search, and LLM-powered workflows — engineered for cost and reliability, not just the demo.

  • Claude / OpenAI integration with tiered model routing
  • Prompt caching and token budgets that keep costs predictable
  • Evaluation and guardrails before it ever meets a customer
  • Built one for my own storefront — it answers real support tickets

Full builds, AI-accelerated

MVPs, internal tools, and storefronts delivered in weeks. AI tooling handles the boilerplate; senior engineering handles the architecture, security, and edge cases.

  • Scoped in a one-hour call, quoted in writing
  • Web apps on Astro / Cloudflare; systems work in C# / .NET
  • Payments, auth, and data handled by hand
  • Deployed to your infrastructure, documented for handoff
Most requested

AI-built codebase rescue

Your team shipped something fast with Cursor, Claude Code, or Lovable — and now it works until it doesn't. I take it over, find out what it actually does, and make it safe to change.

  • Full read-through with a written risk report
  • Auth, payments, and data handling audited first
  • Tests around the parts you can't afford to break
  • Handoff documentation so the next person isn't guessing

Automation & integrations

The manual process that eats a day a week, turned into software. APIs stitched together, workflows automated, bots that do the boring parts.

  • Discord bots, webhooks, and third-party API integration
  • Internal dashboards with a login and a database
  • Specialty: game server engineering (Rust / Minecraft, since 2021)
  • Runs on your infrastructure or fully managed on mine

Case studies

Shipped, sold, and supported

Not client logos — my own products, built end to end and answering to real customers. The strongest proof a freelancer can offer: things that survive contact with users.

AI in production · E-commerce

WarpedBed — storefront with an LLM support agent

A storefront for 3D-printed tabletop gaming accessories, built from an empty repository: catalog, checkout, order pipeline, and fulfillment tooling. Its multimodal support chatbot uses tiered model routing — cheap models for simple questions, capable ones for hard cases — with prompt caching to keep per-conversation cost low. The unglamorous milestone: it passed Google Merchant Center policy review, the gate most small storefronts fail.

Stack
Astro · Cloudflare Pages · D1 · PayPal
AI
Multimodal chat, model routing, prompt caching
Scope
Design → build → deploy → operate
Status
Live, taking orders

Commercial software

Rust server plugins — a product line with paying customers

A catalog of C# plugins sold to game server operators on Codefling: an emergency response event system, a PvP war mode, drone events, container stacking. This is product work in miniature — real customers, real bug reports, and a real update cadence against a game that patches monthly. It's where I learned that shipping is 20% of the job and supporting is the other 80%.

Stack
C# · Oxide / Carbon · Discord API · RCON
Scope
Product design through customer support
Status
Paying customers since 2021

AI in production · Desktop + browser extension

Fiscal Month Tracker — an AI-assisted productivity suite

A commercial desktop application and companion browser extension for customer service operations teams. The extension detects and times chat sessions inside ServiceNow automatically — multi-chat, with per-case isolation so sessions never contaminate each other — while the desktop hub handles fiscal-period analytics: QA scoring, productivity rates, and performance trending against configurable targets. AI response suggestions run on OpenAI and Anthropic models with PII redaction and department-scoped context, and session summaries generate scored, sentiment-aware coaching feedback. It ships with its own update server: background downloads, SHA-256 integrity verification, mandatory-update enforcement.

Stack
Windows desktop · Chrome/Edge extension (MV3) · ServiceNow
AI
OpenAI + Anthropic suggestions · PII redaction · coaching summaries
Scope
Product design → licensing → self-hosted update server
Status
Commercial, licensed

Infrastructure

Production infrastructure I run myself

A single-box production environment hosting game server orchestration, file sync, storefront tooling, and tunneled public access — plus a dedicated Polygon blockchain node (Heimdall + Bor on 7.4 TB RAID0) with custom monitoring. Maintained through the parts nobody puts in a portfolio: JVM segfaults, memory diagnostics, sync recovery, and mod conflicts at 1 a.m. When I say I'll operate what I build for you, this is the evidence.

Stack
Linux · Docker · Pterodactyl · Cloudflare Tunnel
Node
Heimdall V2 · Bor · WebSocket status feed
Status
Continuous multi-year uptime

Desktop tool

DeckPress

A C# tool that pulls card data from the Scryfall API, lays out print sheets, and emits cut files for a Silhouette Cameo — including color-profile calibration for consistent output on consumer printers.

Open source

Panda2Prusa

A Bambu Studio → PrusaSlicer 3MF converter that preserves transforms, multicolor painting, and per-part extruder assignments — solving a real interoperability gap between slicer ecosystems.

github.com/CupsOhJoe/Panda2Prusa →

Internal tooling

Case work system

Case-tracking tooling for service operations — a sibling to the Fiscal Month Tracker suite, built for teams whose work arrives as cases rather than chats.

More on GitHub.

Process

Built like an engineering engagement, not a gig

AI makes the build fast. Process is what makes it safe. Every project runs the same four stages.

01

Scope

One call to understand the problem, then a written scope with a fixed price and a date. If what you need is bigger than what you described, you hear it now — not halfway through.

02

Build

AI tooling for velocity, senior review for correctness. Auth, payments, and data handling are written and audited by hand. Tests go around whatever can't afford to break.

03

Show

Working deployments over status updates. You see something rough running in week one rather than hearing about something perfect in week four.

04

Hand off

Your repo, your accounts, your infrastructure — documented well enough that hiring me again is a choice, not a requirement.

About

Why AI tooling plus a senior engineer

Anyone can generate code now. The market is full of projects built in a weekend that nobody — including their owners — can debug, secure, or extend. The gap isn't generation anymore; it's judgment: knowing what the AI got wrong, what it silently skipped, and what will fall over under real users.

That judgment is what I sell. My background is C# and full-stack systems, and the last several years have been spent shipping my own commercial products — a storefront, a plugin catalog with paying customers, production infrastructure I keep alive myself. When I use Claude Code or Cursor to move fast, I can read every line that comes out, because I've written those systems by hand.

You get AI-tooling speed with an engineer accountable for the result — one person, senior end to end, from architecture to the support ticket six months later.

Core stack

C# / .NET · TypeScript · Astro · SQL

Certifications

Foundational C# with Microsoft ↗

AI tooling

Claude API · Claude Code · OpenAI · model routing

Infrastructure

Linux · Cloudflare · Docker · self-hosted

Availability

10–15 hrs/wk · response within a day

Location

Spokane Valley, WA · working remote

FAQ

Questions clients actually ask

If AI writes the code, what am I paying you for?

The 20% the AI can't do — which happens to be the part that sinks projects. Architecture decisions, security review, cost control, edge cases, and knowing when generated code is confidently wrong. AI lowers my hours, which lowers your price; it doesn't lower the standard the code is held to.

How does pricing work?

Fixed scope, fixed price, in writing, before I start. One call to understand the problem, then a quote with a delivery date. No hourly meters, no surprise invoices. If the scope grows mid-project, we re-scope explicitly — you'll never discover a cost overrun after the fact.

Who owns the code?

You do, from the first commit. The repository lives in your account, deployments go to your infrastructure (or mine, if you'd rather not manage it), and every credential is yours. The handoff documentation is written so that any competent developer can pick it up — hiring me again should be a choice, not a hostage situation.

Can you fix a project another AI tool (or freelancer) built?

This is my most-requested service. I start with a full read-through and a written risk report — auth, payments, and data handling first — then stabilize with tests around whatever can't afford to break. You get an honest assessment of whether it's salvageable before you spend anything on rebuilding.

What do you not take on?

Anything I can't do well part-time: large teams to coordinate, on-call SLAs measured in minutes, or projects that need 40 hours a week. I work 10–15 focused hours weekly, which suits fixed-scope builds, rescues, and features — and I'll say so on the first call if your project needs more than that.

What happens after launch?

Every project ships with documentation and a stabilization window for defects. After that, you can run it yourself with the docs, or keep me on a small monthly retainer for updates and support. I run my own products and infrastructure year-round — maintenance is a first-class part of the offer, not an afterthought.

Contact

Tell me what you're trying to build

Or what's broken. A short description of the problem and your timeline is enough — I'll reply within a day with an honest read on whether I'm the right person for it.

hire@jkpsoftware.com github.com/CupsOhJoe

Message sent

Thanks — it's in my inbox. You'll hear back within a day at the address you gave.

Start the conversation