Transforming Legal Tech with Tax-Fin-Lex

Since January 2023, we’ve been working with Tax-Fin-Lex — a legal and regulatory information company based in Ljubljana, Slovenia that serves all Slovenian courts, government ministries, and nearly every law office in the country. What started as a contract engagement quickly grew into one of the most impactful engineering partnerships in our portfolio.

Here’s a look at the key projects we’ve delivered and the lessons learned along the way.

Building an Internal CRM from Scratch

When we joined, the sales and administration team relied on spreadsheets, emails, and manual processes to manage clients, billing, and contracts. It worked — but it didn’t scale.

We architected and delivered a full internal CRM system built on top of the existing data infrastructure. The system supports:

Sales pipeline management — tracking leads through the entire lifecycle
Administration workflows — automating repetitive tasks around contracts and client management
Billing integration — generating and tracking invoices connected to client accounts

The key challenge was building on top of legacy data infrastructure without disrupting existing operations. By designing the CRM as a layer over the existing database, we avoided a risky migration while still delivering a modern, fast interface.

TFL AI: A RAG Pipeline Processing 50M+ Vectors

One of the most technically ambitious projects was TFL AI — an AI-powered document processing system built on Retrieval-Augmented Generation (RAG) architecture.

Tax-Fin-Lex maintains an enormous corpus of legal documents, regulations, and tax guidelines. The goal was to make this knowledge base queryable through natural language — allowing users to ask questions and get accurate, source-cited answers.

The Technical Stack

Vector database handling 50M+ vectors for semantic search
Document ingestion pipeline that processes, chunks, and embeds legal text
On-premise hardware — all processing runs on local infrastructure for data sovereignty and compliance
RAG architecture combining retrieval with LLM generation for accurate, grounded responses

Building this on on-premise hardware (rather than cloud APIs) added significant complexity around resource management, model serving, and scaling — but it was essential for the sensitive nature of legal data.

You can try TFL AI yourself at tax-fin-lex.si/tfl-ai.

Nexus: Centralised Data Ingestion Platform

Before Nexus, internal data jobs and recurring processes were scattered across cron jobs, scripts, and manual procedures with no central visibility.

We designed and built Nexus — a centralised data ingestion platform that consolidates nearly all internal jobs and recurring processes into a single, observable system. This gave the team:

Full visibility into what’s running, when, and whether it succeeded
Centralised scheduling and monitoring for all data pipelines
Error handling and alerting to catch failures early
A unified interface replacing dozens of fragmented scripts

Saving 100+ Hours Per Week Through Modernisation

Beyond the new systems, a huge part of our impact came from modernising existing workflows:

Manual to human confirmation — We transitioned key workflows from fully manual data entry to automated processes that only require human confirmation. This alone eliminated hours of repetitive work daily.
WPF to web migration — Several critical business tools were built as legacy WPF desktop applications. We migrated these to modern web-based alternatives, making them accessible from anywhere and dramatically easier to maintain.

The combined effect: over 100 hours saved per week across the team.

Engineering Culture Improvements

Technology alone isn’t enough — we also focused on improving the engineering practices of the team:

Git adoption — Standardised version control by establishing Git workflows across the team, replacing ad-hoc file sharing and manual versioning
CI/CD pipelines — Built automated testing and deployment pipelines, reducing deployment risk and speeding up releases
Docker containerisation — Introduced containerisation for consistent development and deployment environments
Grafana observability — Set up monitoring dashboards giving the team real-time visibility into application health and performance
Linux production environments — Managed deployment and maintenance across Linux-based production servers

Key Takeaways

Three years working with Tax-Fin-Lex reinforced some core beliefs:

Build on what exists. You don’t always need a greenfield rewrite. The CRM succeeded because it layered on top of existing infrastructure rather than replacing it.
Automation compounds. The 100+ hours/week savings didn’t come from one big project — it came from dozens of small workflow improvements that compound over time.
AI needs infrastructure first. TFL AI wouldn’t have been possible without first building the data ingestion and processing foundations with Nexus.
Engineering culture is a multiplier. Git, CI/CD, Docker, and monitoring aren’t glamorous, but they make everything else faster and safer.

Results

100+ hours per week saved through automation
50M+ vectors processed and indexed for semantic search
Full observability with real-time monitoring and alerting
Containerised infrastructure enabling rapid, reliable deployments
Modernised engineering culture with Git, CI/CD, and Docker adoption

Internal CRM, AI-powered document pipeline processing 50M+ vectors, and a centralised data platform — saving 100+ hours per week.