Build a AI chatbot with Next.js + Claude API + Vercel
Custom-trained chatbots for support, sales, or internal tools, grounded in your own knowledge base.
Not another GPT wrapper
Most AI chatbots ship as a chat widget that just calls OpenAI with a prompt prefix. That works for demos. It fails in production because it hallucinates, cites things that don't exist, and gives different answers to the same question. A real production chatbot needs retrieval-augmented generation (RAG), guardrails, and observability.
What I build
- Knowledge ingestion pipeline — your docs/FAQ/website ingested, chunked, embedded into Supabase pgvector.
- RAG with citations — every answer cites the exact document and line it came from. If retrieval doesn't find a match, the bot says "I don't know" instead of inventing.
- Guardrails — the bot refuses out-of-scope questions (medical, legal, financial advice) by default. You configure what topics it can engage on.
- Conversation logging — every conversation stored for review. Bad answers get flagged, used to improve retrieval.
- Cost guardrails — daily/monthly token budget per workspace. Alerts at 80%. Hard stop at 100% so a runaway loop doesn't produce a $5,000 surprise bill.
Why Claude
For most chatbot use cases, Claude Haiku is fast and cheap enough to feel synchronous, while being significantly better at "I don't know" behavior than smaller models. For complex internal tools or sales chatbots that need reasoning, Claude Sonnet handles it. The choice is part of the scoping conversation.
Deep-SKAI™hospital supply AI
AI demo for hospital supply chains with interactive ROI calculator, persona-based briefing flows, and Replit handoff package.
Read the case study →FAQ
Can it handle conversation history?+
Yes, stored in Supabase per session. We can also do longer-term memory via embedding past conversations and retrieving them as context, though this is rarely needed for support bots.
What happens when the knowledge base changes?+
Re-ingestion is automated. Push new docs to the configured source, and the embedding pipeline updates within minutes.
Do you offer ongoing maintenance?+
Optional retainer at €500/mo for a chatbot — covers tuning, prompt iteration, knowledge updates.
Ready to build a AI chatbot?
Tell me what you're shipping. Reply within 2 hours during EU business hours.