Chatia Space

Inspiration

A Friday Instagram ad changed everything. Someone offered a way to use my PayPal to receive payments without fees. It sounded shady, so I walked away. But we kept talking. He turned out to be a marketing strategist with clients like dentists and clinics. Our conversation drifted to WhatsApp automation. He told me his clients were desperate to respond to customers using real AI, not a clumsy chatbot. He had tried building it himself but didn't have time. I told him, "I can do that." That was the spark. I wanted to give small businesses the 24/7 response power of big companies at a price they could afford.

What it does

ChatIA Space turns a WhatsApp Business number into an AI-powered sales and support team. It responds to customers 24/7 with natural, human-sounding conversations, not rigid chatbot scripts. It qualifies leads with automatic scoring, books appointments, reads payment receipts with OCR, and notifies the business owner when something needs attention. A visual CRM replaces Excel spreadsheets, and a multi-agent system lets businesses assign conversations to different team members with full traceability. Everything is managed from a single web panel.

How we built it

I started with Google AI Studio and Gemini, running everything locally through Ngrok. Phase 2 brought a proper architecture: I used DeepSeek on OpenRouter to design a multi-service backend with Docker, FastAPI, and PostgreSQL. The frontend went through several iterations — first with Qwen 2.5, then DeepSeek V4 with a million-token context window through OpenGo. I went through five accounts (about $5 USD each) and DeepSeek refined both frontend and backend dramatically.

The system grew so complex I created the "City" analogy to manage it: every module became a building, every notification a street, every security layer a police or fire department. That mental model let me design 84 interconnected components without losing track.

The backend runs on a Debian VPS with Docker and Nginx (HTTPS only). The frontend is on Vercel. The domain sits behind Cloudflare.

Challenges we ran into

Meta's ecosystem was the biggest technical maze: embedded signup, phone number registration, business verification. I'm still navigating it. Making the bot sound human was the other big challenge — I ended up writing 10 golden rules in the system prompt instead of fine-tuning. No bullet points, one emoji max, never reveal you're an AI, one question per message. It works.

The deepest fear is an error slipping through. I researched Row Level Security, Fernet encryption, and multi-tenant isolation extensively. The system passes internal checks, but I still want to run a real penetration test to confirm two businesses can't mix their data.

Accomplishments that we're proud of

Having an adult with 15 years of commercial experience ready to use the system as a paying customer. When someone who has managed real businesses for over a decade sees the value and wants to integrate it into their daily operations, it validates that the product solves a genuine market problem.

On the technical side, building a production-ready SaaS with 84 documented components, 53 repairs executed, zero critical bugs, and a full city-map documentation that any developer can understand in 10 minutes.

What we learned

I already ran Minecraft servers, so Linux and VPS were familiar. What changed was learning to express ideas with precision. When you work with AI-assisted coding, the quality of your output depends entirely on the quality of your instructions. I got better at describing exactly what I wanted.

I also deepened my understanding of LLMs: how to structure prompts so the bot keeps a consistent personality, how to design multi-tenant isolation at the database level, and how to sell technology to non-technical business owners with empathy and simplicity.

What's next for Chatia Space

A real pilot with a watch seller in Medellín who also owns four clothing stores. We're meeting Monday, June 1st, to test the system in his business. Testing with someone who lives a completely different life from mine is the only way to know if ChatIA Space truly works.

After that: launch, a partnership with the marketing strategist who sparked this whole journey, and my dad as legal representative. The goal is to turn this from a solo project into a real company that gives small businesses across Latin America the same AI power as big corporations.

Built With

• apscheduler
• bash
• docker
• docker-compose
• fastapi
• fernet-(encryption)
• github
• google-calendar-api
• google-cloud-(gemini-api)
• groq-api-(llama)
• hetzner-vps
• httpx
• lucide-react
• meta-whatsapp-cloud-api
• next.js-16.2
• nginx
• oauth2/jwt
• openai-api
• postgresql-15
• powershell
• pydantic
• python
• radix-ui
• react-19
• recharts
• redis-7
• sql
• sqlalchemy-(async)
• tailwind-css
• tanstack-react-query
• tanstack-table
• traefik
• typescript
• vercel
• websocket