01 / Project
AI Customer Support Agent
A RAG-powered customer support system that answers customer questions using documentation, service information and operational data.
- Python
- FastAPI
- LLMs
- RAG
- Vector Search
- Problem
- Customer information is often spread across documentation and internal sources, making accurate answers slow to find.
- Approach
- Structure the source material for retrieval, define clear system boundaries and evaluate answers against the available evidence.
- Solution
- A FastAPI service combines vector search with an LLM to retrieve relevant context before producing a grounded response.
- Outcome
- A practical foundation for a support assistant that can be tested with real documentation before wider integration.
02 / Project
VPS Infrastructure Assistant
An intelligent agent capable of retrieving infrastructure status and operational information through tool integrations.
- Problem
- Routine infrastructure checks require switching between tools and interpreting low-level system information.
- Approach
- Expose a small set of auditable infrastructure tools and give the agent only the access needed for each task.
- Solution
- An MCP-based assistant translates operational questions into structured tool calls and returns concise, traceable answers.
- Outcome
- A focused prototype for exploring safer, more accessible infrastructure operations without hiding the underlying data.
03 / Project
Business Workflow Automation
Automation connecting existing business tools and APIs to reduce repetitive administrative work.
- Problem
- Important information is repeatedly copied between systems, introducing delays and avoidable errors.
- Approach
- Map the current process, identify reliable integration points and keep exceptions visible to the people responsible.
- Solution
- A set of API-driven workflows moves and validates data between existing tools while preserving clear operational checkpoints.
- Outcome
- A maintainable automation pattern that can be extended as the business process becomes better understood.