Service
AI, Machine Learning & Data
AI and data that stay under your control.
RAG systems over internal documents, custom AI assistants, self-hosted open models (Llama, Mistral, Whisper, Flux) on Docker/Kubernetes, OCR+NLP document processing, demand and sales forecasting, and ETL pipelines with BI dashboards (Metabase, Power BI, Looker Studio) that bring accounting, CRM and store data into one place.
- RAG architecture with vector search and cited sources in answers
- Self-hosted model deployment on Docker/Kubernetes
- Custom assistant/support bot with a training and testing process
- OCR+NLP document pipeline integrated with ERP/accounting
- Forecasting model with an accuracy report and update plan
- ETL pipeline connecting your data sources
- Data warehouse structure and schema
- BI dashboard built around core business metrics
- Web scraping script with monitoring and alerts
Examples
Typical situations and how I'd handle them
Accounting firm
Invoices that read themselves
Problem
Hundreds of purchase invoices a month are typed into the accounting software by hand.
Solution
OCR and a language model extract the fields and send them straight into accounting: humans only review the exceptions.
Result
Dozens of hours a month back and fewer typos.
Law firm
Private AI on your own server
Problem
Sensitive documents can't go to a public AI cloud, but searching them takes hours.
Solution
A language model running on the firm's own server that searches, compares and summarises its documents.
Result
The benefits of AI without handing over your data.
What's included
What I do as part of this service
RAG over your documents, not the public cloud
Internal wikis, manuals and contracts become searchable without sending data to a third party.
Self-hosted models on your own server
Llama, Mistral, Whisper and Flux on Docker/Kubernetes: you own the infrastructure, you don't rent the API.
Automated invoice and contract processing
OCR + NLP pulls data straight from documents into your accounting or ERP system.
A support bot that knows your product
Trained on your documentation and domain terminology, not a generic ChatGPT wrapper.
Demand and sales forecasting
A model built on your historical data that gives you real numbers, not just trend lines.
An ETL pipeline that runs itself
Accounting, CRM and your store feed one warehouse automatically, no manual export-import.
A BI dashboard that answers real questions
Metabase, Power BI or Looker Studio set up around your business's actual metrics, not a generic template.
Web scraping with monitoring built in
Prices, product listings or real-estate listings collected reliably, with alerts when something changes.
A warehouse that scales with your data
Structured so a million rows doesn't make the dashboard crawl.
FAQ
Frequently asked questions
Does our data go to a public cloud like OpenAI's?
Not unless you want it to. Self-hosted models run on your own server or a cloud of your choosing: data never leaves your control.
Is a self-hosted model as good as a large model like ChatGPT?
For narrower tasks (support, document search, classification) a fine-tuned smaller model often matches it, at a much lower ongoing cost.
How long does building a RAG system take?
A first working prototype usually takes 2–3 weeks; a production-ready version with monitoring and feedback loops takes 6–8 weeks.
What data sources can you connect?
Most systems with an API or database access: accounting software, CRM, e-commerce platforms, Google Analytics and others.
Can the dashboard be adjusted after launch?
Yes, dashboards are built so new views and metrics can be added without rebuilding the whole system.
Is web scraping legal?
Data collection is done in line with each site’s terms and applicable law: I assess what's permitted and sensible case by case.