LLM application architecture
How models interact with APIs, structured data, application logic and production systems — not a chat box as the product.
AI Solutions Architect & Lead Engineer
Yannis Raftopoulos
Founder of RankQuest
18+ years designing and building production software systems. Today I apply that experience to AI-powered applications, decision systems and automation — combining deterministic software with LLM-assisted reasoning. Open to roles focused on LLM and agentic applications.
AI ArchitectureLLM SystemsAgentic AIBackend ArchitectureTechnical Leadership
Scroll
About
18+ years building software systems.
Seniority first. AI is the current application of a long engineering career — not the start of one.
GreeceDubaiAustriaSaaSIoTAI
Over the last 18+ years I have designed, built and shipped production software: architecture, backends, clients, integrations and products — across startups, client systems and my own studio.
That work spans SaaS, automation, IoT and smart-building platforms, hospitality systems, commercial WordPress products, mobile and desktop applications, and more recently AI-powered products using LLM APIs inside real software systems.
I have helped launch mobile games in Greece, worked on smart building and energy management platforms in Dubai, and built hospitality automation and reservation systems in Austria. Some products worked. Some startups failed. The through-line is systems that have to run.
People rarely lack information. They lack clarity.
Founder work is part of the architecture story: taking an ambiguous business problem, designing a system, making trade-offs, and shipping. That is the same muscle required of an AI solutions architect.
- Software architecture
- Full-stack & backend
- SaaS platforms
- Automation
- IoT / smart buildings
- Product architecture
- Technical leadership
- Startup / product development
- GenAI & LLM applications
AI Architecture
What I actually design
AI belongs inside software architecture: data, rules, models, validation and action. The model is a component — not the system.
Agentic systems & workflows
Systems that inspect context, choose a next action and execute a workflow, while software still owns data, constraints and outcomes.
AI decision systems
Combining structured signals, deterministic evaluation and LLM reasoning. RankQuest: evaluate website signals, rank impact, then guide implementation.
AI automation
Putting model calls inside real product flows — analysis, generation and guidance — instead of treating AI as a standalone feature.
Production software architecture
Backends, APIs, databases, data flows and integrations across SaaS, automation and IoT — including event-driven platforms and multi-surface clients.
Technical leadership
Translating business requirements into architecture and an implementation plan, then owning the path from design through shipping.
Flagship case study
RankQuest
AI-powered SEO decision & implementation system
Problem
Traditional SEO tools present large amounts of data. Teams still have to decide what matters and what to do next. RankQuest is built around that gap — between analysis and execution.
Architecture
RankQuest continuously evaluates website signals and identifies the highest-impact growth opportunities. Each recommendation includes implementation guidance and actionable next steps. Current capabilities include technical SEO analysis, content growth planning, SERP-informed content generation and continuous growth monitoring.
AI vs deterministic logic
Website signals, technical analysis and growth monitoring are data-driven. Opportunity identification is product logic: evaluate signals, rank impact, recommend a next step. OpenAI is in the stack for AI-assisted work such as SERP-informed content generation and implementation guidance. Humans remain in the loop for execution.
System flow
Data-driven
Website & search signals
Data-driven
Technical SEO analysis
Deterministic
Signal evaluation
Deterministic
Opportunity prioritization
AI / LLM-assisted
SERP-informed content generation
AI / LLM-assisted
Implementation guidance
Human in the loop
Human or assisted execution
Data-driven
Continuous growth monitoring
Engineering decisions
Deterministic before probabilistic
Website signals are evaluated and opportunities are ranked in product logic first. OpenAI is used where generation and guidance help — not as a substitute for that evaluation.
Decision engine, not a dashboard
The product issues prioritized next actions with implementation guidance, rather than leaving teams inside a pile of SEO reports.
Human in the loop
Execution stays with the user. The system recommends and guides; it does not silently change a live website.
Implementation & status
In active development as my current focus. No public outcome metrics are listed here because none are published on this site.
Next.jsSupabaseTailwindOpenAIVercel
Experience
An 18+ year path through real systems
Roles from LinkedIn, in reverse chronological order — architecture, product, and engineering first, with earlier work included for completeness.
Founder / AI Solutions Architect / Lead Engineer
RankQuest · Full-time
Jan 2026 — Present · Greece · Remote
Building and architecting an AI-powered SEO decision system that turns website and search data into prioritized actions and implementation guidance.
- →Designed the end-to-end architecture combining data-driven analysis, deterministic opportunity ranking and LLM-assisted reasoning/generation.
- →Designed the system so deterministic software handles known rules and prioritization while LLMs are used where interpretation, generation or contextual reasoning is required.
- →Integrated website analysis, external data sources and AI workflows into a single decision pipeline.
- →Owned architecture and hands-on implementation across backend, frontend, database, AI integration and production deployment (Next.js, Supabase, OpenAI, Vercel).
- →Defined the product architecture around one prioritized next action rather than dashboards of raw metrics.
Co-Founder / Software Architect / Senior Developer
Keyonic GMBH · Contract
Jan 2022 — Jun 2026 · Vienna, Austria · Hybrid
Co-leading the development of a smart automation platform, designing scalable architecture and driving technical strategy.
- →Designed scalable architecture and defined technical direction for a smart automation platform.
- →Co-led full-cycle product development from architecture through implementation and market readiness.
- →Led quality standards, feature delivery and technical strategy alongside hands-on development.
Software Architect / Senior Developer
JoyFoundry LTD · Contract
Jan 2014 — Jan 2022 · Athens, Greece · On-site
Architected and developed scalable enterprise applications across backend and frontend systems.
- →Designed and built scalable enterprise applications across backend and frontend.
- →Set technical standards and owned architecture decisions on projects that required non-standard solutions.
Senior Developer / Development Team Leader
Neuhaus Media · Freelance
Jan 2013 — Jan 2015 · Zürich, Switzerland · Remote
Led a team building a full-stack virtual magazine application for iPad.
- →Led a team delivering a full-stack virtual magazine for iPad.
- →Owned project lifecycle and backend/frontend integration through to a high-performing digital publication.
Mobile and GIS Developer
Trans Adriatic Pipeline (TAP) AG · Full-time
Jan 2013 — Jan 2014 · Athens, Greece · On-site
Developed a Windows-based GIS mobile application for collecting cadastral data used in infrastructure planning.
- →Built and integrated a Windows-based GIS mobile application for cadastral data collection.
- →Delivered a geospatial field tool to improve data accuracy and accessibility (Windows Mobile, ArcGIS Mobile).
Project Manager / Software Developer
NDiastasi · Freelance
Jan 2011 — Jan 2013 · Athens, Greece · On-site
Managed software development projects from planning through deployment.
- →Led software projects from planning through deployment.
- →Directed resources and development to deliver applications aligned with client objectives.
Mobile App Developer / Web Designer / Web Developer
VERIAH LIMITED · Full-time
Jan 2009 — Jan 2011 · Athens, Greece · On-site
Designed and developed mobile and web applications, with emphasis on usability and performance.
- →Built interactive platforms that met client specifications and delivered an engaging user experience.
Web Designer
Eurobank · Full-time
Jan 2009 — Oct 2009 · Athens, Greece · Hybrid
Enhanced the design of one of Eurobank’s platforms with CSS and HTML improvements for visual consistency and navigation.
- →Worked with internal teams on branding and refined interface design.
Web Designer / Web Developer
Inaccess by Power Factors · Full-time
Jan 2008 — Jan 2009 · Athens, Greece · On-site
Developed and designed the frontend for smart home and solar management platforms.
- →Focused on interfaces that supported key platform functionality and clearer user interaction.
Call Center Representative
Germanos · Part-time
Jan 2007 — Jan 2008 · Athens, Greece · On-site
B2C sales over the phone, focused on product promotions and client acquisition.
- →Worked to sales targets through product knowledge and customer communication.
Selected systems
Architecture in production
Systems where several moving pieces had to work together — IoT, AI-assisted product flow, and a long-lived commercial product.
Litcom Smart Building Automation
IoT platform across mobile, web and backend
- Role
- Software Architect · Senior Developer
- System
- Mobile + Web + Backend + Realtime + IoT
Problem. Building operators need one system to control devices, track events, program scenarios and receive notifications — across protocols, not a pile of isolated apps.
Built. A complete automation and management platform: device control, event tracking, scenario programming and notifications, including KNX.
Architecture. Multi-surface architecture: Quasar clients, Unity 3D, Node.js backend, MariaDB, Python, WebSockets and KNX. Event-driven control rather than a single-page UI talking to devices directly.
Key challenges
- Realtime communication (WebSockets)
- Device control & KNX integrations
- Event tracking and scenario programming
Data-driven
Devices & KNX
Deterministic
Backend services
Deterministic
Events & scenarios
Deterministic
WebSockets
Human in the loop
Mobile / web / Unity
Quasar FrameworkUnity 3DNode.jsMariaDBPythonKNXWebSockets
- Role
- Founder · Product Engineer
- System
- Web SaaS + LLM API
Problem. Many founders stall between inspiration and a validated opportunity. Generic chat is not a structured way to generate, evaluate and refine a software business idea.
Built. A platform that combines structured thinking, market context and AI-assisted analysis so founders can generate, evaluate and refine SaaS ideas.
Architecture. Product architecture around a repeatable ideation flow — not an unbounded chatbot. Next.js, Supabase and OpenAI, with the model used inside a defined process.
Key challenges
- Structured ideation instead of generic chat
- Combining market context with AI-assisted analysis
Next.jsSupabaseTailwindShadCN UIOpenAIVercel
- Role
- Founder · Product Engineer
- System
- WordPress + Vue/Quasar + MariaDB
Problem. Site owners needed flexible gallery experiences without assembling a fragile stack of plugins, layouts and performance workarounds.
Built. A WordPress plugin that became a commercial product ecosystem: gallery layouts, image management, optimization and customization.
Architecture. Long-lived product architecture: Quasar/Vue client, WordPress/PHP, MariaDB — plus distribution, support and maintenance over time, not a one-off build.
Key challenges
- Product ecosystem beyond a single plugin
- Distribution, support and long-term maintenance
Quasar FrameworkWordPressVue.jsMariaDBPHP
Architecture philosophy
Software that doesn't just inform.
Software that reasons.
Use AI where judgment is needed. Use deterministic software where certainty is needed.
Rules when the answer is known. Models when interpretation is required. Humans when accountability matters.
Strong AI systems combine the pieces below. AI is part of a well-designed architecture — not a replacement for one.
01
Deterministic software
Data models, APIs, constraints, persistence and execution paths that must be reliable.
02
Structured data & signals
Collect and normalize what the system can actually evaluate — not a dump of raw reports.
03
Business rules
Prioritization and eligibility belong in product logic when the answer should be stable.
04
AI reasoning
Use a model where judgment, generation or interpretation helps — not as a substitute for the rest of the stack.
05
Validation & oversight
Recommendations still need checks. People stay in the loop where execution has real cost.
06
Action
The system should help someone do the next step, not only describe the world.
Skills
Tools I have actually used
Current stack and architecture first. Older technologies sit separately — they are real production history, not what I lead with.
AI Engineering
- LLM APIs (OpenAI)
- AI-assisted workflows
- AI decision systems
- AI-assisted content generation
- Agentic / AI workflows
Backend & Architecture
- Solution architecture
- Software architecture
- SaaS / product architecture
- API & backend architecture
- System integrations
- Event-driven systems (WebSockets, IoT)
- Node.js
- TypeScript / JavaScript
- SQL (MariaDB, MySQL, Supabase)
- Python
- PHP / WordPress
- Docker
Frontend
- React / Next.js
- Vue.js / Quasar
- React Native / Expo
- Electron
- Swift / UIKit
- Android
Product & leadership
- SaaS development
- Product strategy & roadmap
- Technical leadership
- UX / product design
- Project management
Platforms & domains
- Vercel
- Cloudflare
- Neon
- Supabase
- Stripe
- Creem
- WordPress
- iOS
- Unity 3D
- KNX
- GIS / ArcGIS Mobile
- C# / WPF
- C++ / Cocos2D-X
- Java
Foundations
- Mathematics
- Statistics
- Physics
Earlier technologies
Shipped in production. Not the stack I design with today.
- Objective-C
- ASP.NET
- Joomla
- jQuery
- Flash
- ActionScript
Before AI was the product
Selected previous work
Before LLMs became part of my work, I spent more than a decade shipping SaaS, mobile, IoT, games and business systems. Six examples — the rest lives on a separate page.
Own Venture
MindJam
Global Trivia Competition
A knowledge quiz game for iOS and Android where players answer thousands of questions across Science, Geography, History, Sports, Cinema and more. Features image-based questions, f…
Cocos CreatorC++PHPMySQL
Own Venture
Snip-Share
Screen Captures to Social Content
SnipShare transforms screen captures into engaging social media content. The app recognizes images and generates tailored post suggestions for X.com, turning the friction of sharin…
Electron.jsReactTailwind CSSShadCN UIVercel
Client Work
GDMAg
Virtual Men's Magazine for iPad
An interactive virtual magazine for iPad designed for men, offering curated content on lifestyle, fashion, technology, health and culture — delivered through video and interactive …
iOSUIKitSwiftPHPMySQL
Contact
Let's build intelligent systems.
Open to AI Solutions Architect, AI Architect, Lead AI Engineer and Senior AI Engineering roles focused on LLM and agentic applications.