abbysconsult.space: a consulting firm's website that never sleeps

When credibility is the product, the website is the first proof. Abbys Consult is an independent Belgian strategy consulting firm: every visitor lands with a question — can I trust this person with a serious problem? The site has to answer that question in the first five seconds, without a big-agency budget and without requiring the consultant to touch a server.

Design as an argument

The design language is built around restraint: editorial typography (no decorative fonts, no gradients), generous whitespace, a strict grid and a muted palette. Credibility through precision rather than decoration. Every layout decision was audited against one question: does this make the firm feel more or less rigorous?

Architecture: server-rendered, secure, fast

The platform runs on Next.js with full server-side rendering, deployed behind nginx on a hardened VPS. Security headers (CSP, HSTS, X-Frame-Options) are set at the nginx layer. Structured data (Organization, Person, Service in JSON-LD) is injected server-side. Time to first byte measured in production: under 200 ms. Technical SEO issues at launch: zero. Lighthouse score on first audit: 98.

The AI assistant: NVIDIA NIM on the backend

Beyond the showcase, abbysconsult.space runs an AI assistant backed by a Node.js/Express API deployed on the same VPS. The assistant uses NVIDIA NIM inference (meta/llama-3.1-70b-instruct) to answer questions about the firm's services, methodology and positioning — with a system prompt that encodes Abbys Consult's exact voice and scope. The backend handles session context, rate limiting and graceful fallback. The consultant updates the knowledge base via a simple content file; the server handles the rest.

Operations: zero client time

Hosting, HTTPS (Let's Encrypt, auto-renewed), monitoring, uptime alerting and content updates are all handled at the infrastructure level. The consultant has never opened a terminal. Since launch, the platform has had zero incidents requiring manual intervention. That is the point: a professional practice should spend its time on client work, not on IT.

Stack and live URLs

Frontend: Next.js, TailwindCSS, JSON-LD. Backend AI assistant: Node.js, Express, NVIDIA NIM (llama-3.1-70b). Infrastructure: nginx, Let's Encrypt, PM2, VPS Hostinger. Live: abbysconsult.space — api endpoint: abbysconsult.space/api/abbys-chat.

Token cost vs. Belgian human equivalent

A Belgian web agency building a comparable platform — editorial design, server-rendered architecture, structured data, security audit, AI assistant integration — typically quotes €6,000–12,000 for the initial build with a 4–8 week turnaround. Monthly managed hosting and maintenance: €300–600. A part-time virtual assistant or receptionist covering visitor inquiries costs €1,200–2,000/month gross. Total first-year cost with a traditional vendor: €25,000–43,000. The AI infrastructure: NVIDIA NIM llama-3.1-70b-instruct inference for approximately 200 visitor conversations/month at €0.77/M tokens ≈ €0.08 in token costs (essentially zero at this volume); VPS hosting shared across projects ≈ €4/month allocated cost. Total monthly AI cost: under €10. Build cost saved vs. agency: €6,000–11,000. Monthly ongoing savings vs. managed hosting plus part-time assistant: €1,490–2,590. Over 12 months, the platform generates an estimated €24,000–37,000 in operational savings against under €120 in AI infrastructure spend.