Frontend
Fast pages and clear interfaces
SEO-ready websites, landing pages and chat interfaces with clean structure and strong performance.- Astro
- Next.js
- TypeScript
- JavaScript
- HTML
- CSS
- Playwright
Engineering stack
Technologies for controlled AI
Useful AI systems are not built from a model alone. They need clear data paths, robust backends, visible sources, tested prompts, secure integrations and clean handoff to humans.
Technologies
Technology for real workflows
We choose technology based on process, risk and data. Not every project needs GraphRAG or fine-tuning. Every project does need clean interfaces, tests, monitoring and reliable handoff.
View technologies Input Routing Knowledge Action Handoff
Backend & data
Interfaces, validation and persistence
Server-side logic for forms, CRM flows, webhooks, databases and secure provider integrations.- Kotlin
- Python
- Node.js
- PostgreSQL
- SQL
- Supabase
- REST
- JSON Schema
AI & retrieval
Answers from documents and context
RAG architectures for document-grounded assistants with sources, boundaries and quality checks.- RAG
- GraphRAG
- LangChain
- Embeddings
- Vector Search
- Hybrid Search
- Reranking
- Fine-tuning
- Evals
Automation
Actions instead of just chat
Controlled automations with tool calling, MCP servers and handoff to humans or business systems.- MCP
- Tool Calling
- n8n
- Telegram
- CRM
- Webhooks
- Audit Logs
Principles
Start lean, scale cleanly
Process before model
We clarify task, data and risk before deciding on agents, RAG or fine-tuning.
Traceable answers
Document-grounded answers need sources, boundaries and test cases. Critical commitments stay with humans.
Integration over isolated demos
An assistant must fit into websites, CRM, messengers and internal workflows, otherwise it remains a nice experiment.
Technology follows the workflow
We choose tools after data, risk, handoff and operations are clear. That keeps the project lean, testable and ready to extend.