RAG
AI That Answers From Your Documents
A general model knows the internet and nothing about your contracts, your SOPs or your price list. RAG closes that gap: your documents are indexed, the relevant passages are retrieved for each question, and the answer is generated from them - with the source shown, so a person can check it.
Key takeaways
- RAG is retrieval before generation. Most bad answers are retrieval failures, not model failures.
- How you split the documents decides the quality more than which model you pick.
- Showing the source passage is not a nicety - it is what makes the answer checkable and therefore usable.
- Fine-tuning teaches style, RAG supplies facts. Businesses almost always need the second one.
What I build
- Ingestion from PDFs, docs, spreadsheets, your site and your CRM
- Chunking and embedding tuned to your document shape
- Vector search with metadata filtering and permission awareness
- Answers that cite the passage they came from
- Re-indexing when the documents change, so answers do not go stale
- An evaluation set so you can tell whether a change made it better or worse
How it works
A short scoping call
What hurts today, what you already use, and what "done" looks like.
A written fixed-price quote
Scope, price and timeline in writing, approved before any work starts.
Built in stages, on your real data
You see a working first version early, not a demo at the end.
Handover with documentation and monitoring
You own the code and the data, and a support window follows launch.
Working with me from the US or Europe
- Most of my client work is for companies in the US, UK, Germany, the Netherlands and Australia.
- Calls are in English and scheduled in your time zone. My working day covers most of the European one and the US East Coast morning.
- A fixed price in US dollars, agreed in writing before any work starts.
- Pay me directly, or hire me through Upwork if you prefer to work through a platform.
- You own the code and the data, with documentation and a support window after launch.
Guides on this
Practical guides from real client work.
Business Workflow Automation with Python: A Practical Guide
A practical, no-hype guide to business workflow automation with Python: which processes to automate, the building blocks, reliability, and a worked example.
Building AI Agents for Business Process Automation: A Practical 2026 Guide
What AI agents for business process automation actually are, where they help, where rules win, plus architecture, costs, and a real adoption path.
Business Automation for Small Business: What to Automate First
A tactical guide to business automation for small business: how to spot what to automate first, the top areas to start with, and a simple ROI framework.
How to Improve Business Efficiency: A Practical Playbook
A practical playbook on how to improve business efficiency: map the work, cut waste before you automate, standardize, pick the right lever, and measure results.
Zapier vs custom code: when no-code automation stops paying off
Zapier vs custom code, decided honestly. Where no-code wins, where per-task pricing and limits hurt, and why AI-assisted code is now viable for small teams.
Make.com vs custom code: where the visual builder hits its limits
Make.com vs custom code, from real builds: where the visual builder shines, the walls it hits at scale, and when going to code is the faster, cheaper call.
n8n vs Custom Code: Even the Developer-Friendly Automation Tool Has Limits
n8n is the most developer-friendly no-code tool out there, and I use it often. But at real complexity, custom code wins. Here is exactly where the line sits.
How Much Does Business Automation Cost? (2026 Pricing Guide)
How much does business automation cost in 2026? A clear price guide by automation type, one-time build vs ongoing fees, what drives the price, and how to think about ROI and payback.
Should You Hire a Freelancer to Automate Your Business?
Should you hire a freelancer to automate your business? When DIY no-code is enough, when to hire, freelancer vs agency vs in-house, the red flags to avoid, and how to scope a first project.
Frequently asked questions
Is my data used to train a model?
Not in this design. The documents stay in your own database and are retrieved at question time; the model sees only the passages needed for that one answer. Which provider processes those passages, and under what terms, is a decision we make explicitly rather than by default.
RAG or fine-tuning?
Fine-tuning changes how the model writes. RAG changes what it knows. If your problem is "it does not know our prices and policies", that is RAG. If it is "it answers correctly but in the wrong tone", that is closer to fine-tuning, and usually a prompt fixes it more cheaply than either.
How do you know it is actually working?
With an evaluation set: real questions with known correct answers, run after every change. Without one you are guessing, and every model swap becomes a coin flip. This is the part most RAG projects skip and then cannot debug.
Sources
Official documentation you can verify the claims on this page against.
Let us scope your project
Tell me what you need and I will give you an honest scope, price, and timeline.