AI
Agents
Automate recurring decisions and tasks with access to internal systems.
Pilicore Agent
How I think and build solutions
The Pilicore-Agent is the reference architecture behind how I work. Not a product to buy, but proof of how I build AI and automation solutions in a modular, extensible way without vendor lock-in.
That is how I work in client projects: process problem first, then the right architecture. The screenshots show the current state as evidence of that thinking.
Tech stack for those interested
Next.js, FastAPI, PostgreSQL + pgvector, Celery, LiteLLM, Qdrant, Docker.
What this helps solve
FX
Workflows
Connect systems and processes without media breaks or manual handoffs.
KB
Knowledge & chat
Answers based on your documents and company data, where the team actually works.
API
Integrations
Connect existing tools instead of replacing the landscape. Data flows, ownership stays clear.
Use case
Invoice intake from email
Business problem
Invoices arrive by email. Attachments must be detected, checked and filed. Done by hand, that costs time and creates errors.
Workflow
A flow detects attachments, processes structured formats automatically, flags unclear PDFs and writes to SharePoint.
Result
Less manual work, traceable filing, clear status. The business logic stays, even if tools or models change.
Typical process problems
Not detailed client cases. Real business problems that the same way of thinking can solve: process first, technology second.
Applicant management
Problem: applications sit scattered across inboxes and folders. Result: structured capture, routing and status without media breaks.
Contract review
Problem: contracts must be checked for clauses and deviations. Result: an initial reading and handoff to the right person.
Knowledge search
Problem: people search for information instead of working. Result: answers based on internal documents and sources.
Customer service
Problem: recurring requests tie up specialists. Result: pre-qualification and relief, with clear escalation to people.
Quote handling
Problem: quotes are built from email, Excel and copy-paste. Result: pull data together, prepare a draft, trigger approval.
Complaints
Problem: complaints get lost or stuck in inboxes. Result: capture, prioritise and route to the right team.
Approvals
Problem: approvals run through chat and email with no overview. Result: a clear flow, status and a traceable decision.
Quality management
Problem: checks and deviations are documented by hand. Result: standardised capture and handoff to follow-up steps.
Documentation
Problem: minutes, reports and filing happen late or not at all. Result: generate and file them from existing data.
Onboarding
Problem: access, checklists and ramp-up depend on individuals. Result: a repeatable flow across systems and owners.
Orders
Problem: orders are copied between mail, ERP and Excel. Result: handoff without media breaks and with clear status.
A look at the interface
Architecture principle
Business logic stays in place.
Only the connections change. LLM providers, tools and integrations are swappable. So your solution stays workable when prices, vendors or models change.
If you have a comparable process problem, I will talk with you about your use case. Not about the platform.