Filera — applied AI engineering
We talk to your team, map the manual work, and build the AI system that replaces it — inside your real infrastructure.
Our approach
Yes, we run a discovery. No, we don’t write 40-page reports. We talk to the people doing the manual work, map it, and build the system that replaces it.
A short call (or two) with the people who actually do the work — finance, ops, sales, whoever owns the manual process. We listen, ask, and write down what’s actually happening, not what the org chart says. Remote, no on-site week required.
We turn the call notes into a concrete map: inputs, outputs, decisions, edge cases, where humans are doing pattern-matching a model can do, and where they’re doing judgment a model can’t. The map becomes the contract for what we build.
We build inside your infrastructure — your cloud, your data, your auth. Production code, version-controlled, monitored, observable. Not a notebook, not a demo. The first version goes live in weeks, not quarters.
We run the system with you while it bakes in — handling edge cases, tuning prompts, watching drift. Then we hand it off to your team, cleanly. No zip file, no “good luck.”
What we build
Pipelines that turn unstructured documents — PDFs, scans, contracts, emails — into structured data your systems can actually use. OCR plus LLM extraction, validated against a domain schema. Shipped for real estate, finance, and ops teams.
Replace manual operations inside existing systems — SAP, ERPs, CRMs, internal tools — with LLM agents that operate the UI directly, with a human approval queue where it matters. Production-grade orchestration, not a notebook with prompts.
Full automation of multi-step workflows from input signal to outcome. Recruiting fulfillment from sourcing through scheduled call. Customer-support triage. Internal ops pipelines. Built to run unattended, with audit trails.
In production
No mockups, no concept videos. Each of these is running today.
Problem
Outbound sourcing, candidate evaluation, personalized outreach in two languages, reply triage, scheduling, candidate document generation. Normally 30+ hours per recruiter per week.
System
End-to-end pipeline: pulls jobs from the CRM, runs sourcing via Xing TalentManager + LinkedIn, evaluates candidates with a calibrated LLM scout, drafts personalized German + English outreach, triages replies, books calls, generates per-client candidate documents.
Status
In production at Filera’s own recruiting business. Also deployed for partner recruiting agencies — Bricklog and Leithold GmbH. One operator, multiple agencies plugging into the same stack.
Problem
Analysts manually extracting structured data — terms, rent rolls, expenses, options — from lease packets and real-estate documents. Hours per document.
System
Document-intelligence pipeline: OCR plus structured extraction with an LLM against a domain schema, with validation and a review queue. Output ready for downstream financial modelling.
Status
In production with the real-estate practice at Horváth & Partners. Documents processed monthly, structured output feeding the practice’s internal models.
Problem
Customer-service team manually performing repetitive operations inside an ERP — order changes, status lookups, escalation routing. Bottleneck on growth.
System
LLM agents operating the ERP UI directly via Playwright plus LLM orchestration. Human approval queue for non-routine actions. Full audit trail.
Status
Active client engagement. Routine tickets handled without human touch; complex cases routed to humans with full context.
About
Filera is Finn Radoy — a German applied-AI engineer based in Malaysia. Over the last 18 months: document-intelligence pipelines for the real-estate practice at Horváth & Partners, ERP-automation work for a European logistics-software company, and a fully-automated recruiting platform I run for my own agency and for partner agencies including Bricklog and Leithold GmbH.
Filera is the consulting vehicle for that work: forward-deployed AI engineering for businesses that want shipped systems, not slide decks. Always remote. Always inside your real infrastructure.
Questions
We don’t sell hours and there is no team of generic engineers behind the brand. Filera is one forward-deployed operator who lives inside your stack until the system ships. The deliverable is a production system, not deliverables-per-sprint.
Get started
Tell us about the manual work. We’ll tell you what we’d build.
