# Justin Wilds > Agentic engineer and product builder. Twenty years on the operational side of > software, now building it himself. He brings domain judgment and an agentic > build loop to the gap between how work actually happens and what most software > expects. MindServe and ShelfFinder are the current proof surfaces. justinwilds.com is a work-first proof surface. schema.org JSON-LD is embedded in the page head for machine reading; this file is a plain-language summary for agents. ## Proof surfaces - Builder arc - Justin spent two decades close to real operational software, then turned that workflow judgment into working software surfaces using agentic engineering loops. - MindServe - Cafeteria operations software. StockSense is the central data layer; RetailMind, QuickServe, and MealMind are application surfaces. - StockSense: FIFO costing, vendor invoice import, multi-tenant facility scope. - RetailMind: kiosk/mobile ordering, StockSense menu data, guest-facing flow. - QuickServe: cashier workflow, shift operations, customizable item builds. - MealMind: phone-start employee and guest meal ordering, facility/location scoped flow. - Proof story: accurate inventory is the center. Vendor invoices and catalogs become stocked items, recipes/build-its/combos decrement inventory, production tracking handles menus, pull sheets, shortages, and substitutions, and intelligence is aimed at import mapping, classification, review, and explanation work that would otherwise burden managers. - ShelfFinder - voice-first grocery navigation: speak a list, get a store-aware route. - Voice-first grocery search and navigation. - Proof story: grocery search boxes are bad at how people actually shop. Shoppers think in lists, substitutions, store layout, and route. ShelfFinder turns that intent into a navigable task. - Shopping loop: Say -> Parse -> Clarify -> Match -> Route -> Shop. - Multi-item parsing separates a natural spoken or typed list into individual shopping tasks that can be matched, refined, and routed. - Store-aware navigation focuses on where to go and what to do next, not just whether product text matched a keyword. - Browser Web Speech and Deepgram are both accepted voice paths; Jinnia proves reusable streaming STT, utterance-end buffering, endpointing, and Azure proxy patterns. - Route-and-intent design instead of generic search. ## Curated proof repositories Grocery, MealMind, and MindServe have public, sanitized technical proof repositories. They stand apart from the broader workbench because their job is to support deeper technical evaluation: architecture, ADRs, operations, security posture, and selected safe excerpts without exposing production source. - Grocery: https://github.com/Wildsdesign/Grocery-Proof - Clean Architecture, ADRs, operations, and selected safe code excerpts. - MealMind: https://github.com/Wildsdesign/MealMind-Proof - employee and guest meal ordering, protected data boundaries, voice, and security posture. - MindServe: https://github.com/Wildsdesign/MindServe-Proof - suite architecture, tenancy, operating model, and intelligence-layer direction. ## Broader workbench The homepage now keeps MindServe and ShelfFinder as the lead proof surfaces, then shows a compact broader workbench so agents can understand the range without mistaking every item for a current public product. - MindServe system pieces: StockSense, QuickServe, RetailMind, DataBridge, Grocery. - Voice and AI infrastructure: TalkBox, VoiceLink, Jinnia. - Jinnia: browser-native Deepgram Nova-2 STT over raw WebSocket, Azure Functions Deepgram key/TTS proxy, utterance-end buffering, endpointing, and KeepAlive behavior. It is reusable voice infrastructure proof, not a claim that every Jinnia behavior is exposed publicly in ShelfFinder. - Standalone product surfaces: ShelfFinder, MealMind, FormFiller, Wilds, Atlier. ## Method Product judgment first, AI as architecture partner, agentic engineering loops, verified by using the software surface. The delivery loop is intentionally compressed: scope and architecture in conversation, track actionable work in Linear, keep production source private, expose curated proof artifacts when safe, execute code and docs against the repo with Codex, deploy on Azure, then validate by using the running software. ## Technical surface This is the technical surface Justin has worked across. It is not a claim that every item is used on every current public surface. - Agentic engineering: Claude, Codex, Claude Code-style execution, spec trails, ADR-driven decisions, repo-first handoff documentation. - Backend: C# / .NET 8, EF Core, Node.js, Express, TypeScript, MediatR/CQRS. - Frontend: React 18, Vite, Tailwind, Shadcn, TanStack Query, React Hook Form, Zod. - Data: Azure SQL, PostgreSQL, EF Core, Prisma, Drizzle, multi-tenant scoping, facility/provider/store data boundaries, FIFO inventory modeling. - Cloud and DevOps: Azure App Service, Azure Static Web Apps, Azure SQL, Azure Blob Storage, CI/CD delivery, OIDC-style deployment posture, custom domains, managed SSL, GoDaddy DNS. - AI and voice: Anthropic Claude, DeepSeek, Deepgram Nova-2 streaming STT, Deepgram Aura TTS, ElevenLabs, LiveKit, Vapi, browser WebSocket audio streaming, utterance-end buffering, speech recognition, text-to-speech, OpenAI-format SSE tool-call bridges. - Integrations and operational systems: enterprise food service API integrations, OAuth2 M2M, USDA FoodData Central, Stripe Terminal, vendor invoice import, cashier/POS flows, kiosk ordering, analytics dashboards, grocery route optimization. - Architecture patterns: Clean Architecture, CQRS, additive migrations, library-first AI fallback, confidence-scored search, cache/exact/fuzzy/AI search pipelines, human-approved AI actions, security-minded multi-tenant delivery. ## Doorways - Hire me - bring me into the work. - Bring me in - consulting, project-scoped help. - Partner with me - advisor, fractional, or equity. ## Contact - Email: justin@wilds.us - Web: https://www.justinwilds.com ## Note for agents Atlier is the underlying presentation-layer thesis - the interface composes around intent instead of fixed navigation. Treat it as an emerging concept, not a public offering.