4xxi
Book a call (opens in a new tab)

We build AI that works when it matters. And the software around it.

Software and AI products for legal, finance and education companies, since 2008. Working software in days, not quarters.

Book a call (opens in a new tab)Start with an audit

Kickoff typically within 3 business days of signing.

  • Engineers with 15–20 years each
  • Founded by a scientist
  • Clients stay 5+ years

This is how our systems answer. Ask one about 4xxi.

Offers (web)Team (web)Terms (PDF)

Question: How soon can you start? Answer: Typically within 3 business days of signing. Source [1]: Offers (web) · § Engagement: “Kickoff typically within 3 business days of signing.”

01Three ways to start

Fixed price. Measured results. Then we keep going.

Start small and see the work before you commit. The audit fee is credited in full toward a project signed within 60 days.

Start here2 weeks

AI opportunity audit

$5–10kfixed

We look at your product and your operations — development, support, marketing, back office — and tell you where AI will actually pay off, where it won’t, and which corners are safe to cut.

  • Where AI helps, and where it doesn’t
  • What to automate first, what to leave alone
  • Your existing AI reviewed, if you have one
  • Measured report + roadmap

Full money-back guarantee. The fee is credited in full toward any project signed within 60 days.

Start with an audit →
Prove it5 business days

Prototype sprint

$20kfixed

A working, deployable prototype on your data, with a fixed scope agreed up front.

  • Runs on your documents and systems
  • Deployable, not a demo
  • Fixed scope and price
Plan a sprint →
Keep goingOngoing

Ongoing engineering

Hourlyrate on request

Senior engineers only. No juniors, no handoffs. Quoted in hours, not seats.

  • The people you meet build it
  • Founder involved
  • Can join your existing team and codebase
Book a call → (opens in a new tab)

Our clients’ products have been covered by

02How we work with AI

How we make AI you can rely on.

We use AI wherever it pays off — to ship features faster, to skip work that no longer needs doing, and to build AI products themselves, like Edtek Chat, Draft and Cite. Here is how we keep it reliable.

  1. 01

    Robust where it must be, probabilistic where it can be

    LLMs are probabilistic. We don’t let them be the only judge. Deterministic checks — scores, rules, assertions — sit alongside model-based evaluation, and the system is built so that a wrong answer is caught, not shipped. The same approach runs through every product we build, not only chatbots.

    deterministic_check() && llm_judge()
  2. 02

    Research, applied

    Semantic chunking, multi-criteria reranking, calibrated LLM-as-judge: the methods come from current papers, not from tutorials. The founder publishes on LLM evaluation in peer-reviewed journals (Scientific Reports (Nature), 2026) and we use the same rigour on client work.

    method ← paper, not tutorial
  3. 03

    Measured before launch, and after

    Before anyone relies on an AI feature, we run it against real cases from your business — documents, tickets, transactions, estimates, whatever the system will actually see. We measure where it is right, where it fails and what it costs. Baselines, regression suites and cost budgets are part of the delivery, so a model change six months later can’t silently break what worked.

    baseline → change → regressions[]
AAAi Chat Book answering a question about the Labor Arbitration Rules, with the cited rule highlighted in the source document

Case · AI · Legal

AAAi Chat Book

AI guidance for the American Arbitration Association. Answers come from the AAA’s own handbooks and rules, each with the passage it was drawn from. Four editions live: Case Preparation and Presentation, Fundamentals of Labor Arbitration, Arbitration for Non-Attorneys, Mediation Practice.

launched
Jan 2025
editions live
4
of queries
1000s
client through
2028
“A breakthrough in making arbitration more efficient, accessible and fair.”
Bridget McCormack, President & CEO, American Arbitration Association
PressLexisNexis (opens in a new tab)LawNext (opens in a new tab)AAA press release (opens in a new tab)
Read the case study →Try the Chat Book ↗ (opens in a new tab)
Our product

Edtek — AI products built on your own content

Edtek Chat, Draft and Cite turn a library a company already owns into AI products: a chatbot that cites by page, a document builder, and inline annotations with the rules and authorities behind every claim. Built for legal teams, publishers and professional services. Used by the American Arbitration Association.

Chata chatbot that cites by pageDrafta document builderCiteinline annotationsedtek.ai (opens in a new tab)
03Track record

AI is our focus now. Everything we learned in 18 years before it is why it works.

Fintech, legal, education and consumer products, shipped since 2008. That experience in delivery and business is what lets us use AI well — and skip the mistakes teams without it make. Filter by what matters to you.

products shipped
100+
raised by our clients
$500M+
average client relationship
5+ yrs
building since
2008
All25AI4Fintech7Legal4Edtech3Other13
All 25 case studies →
“Working with the team has been one of the best commercial experiences of my business life.”
Steven EmanuelPresident, Emanuel Quant Services
“Our team feels that 4xxi is incredibly responsive, dedicated and smart. When something is not clear, they ask the right questions.”
Emily JipsonGeneral Manager, WorthFM
“They are a fantastic team and we wouldn’t be where we are today without them!”
Josh LewisDirector & Co-Founder, Challenge Social
14+ reviews on Clutch ↗ (opens in a new tab)
04Why 4xxi

Seven reasons clients stay for years.

01

We actually care

Your product is our product. If something’s wrong with it, it bothers us at night. No project gets less than full attention — that’s not a slogan, it’s how a small team survives for 18 years.

  1. 02

    Trust is the business model

    We earn it on the first project and keep it for years. Almost every client we have came through another client.

  2. 03

    Founded by a scientist

    PhD in astronomy, still publishing. We work with AI the way a lab works: hypothesis, experiment, measurement. That’s why our systems hold up in production, not only in demos.

  3. 04

    Current stack, not comfortable stack

    LLMs since 2022, ML since 2012, and the best tools available before that. We adopt what works as soon as it works.

  4. 05

    You talk to the people who build

    No sales layer. The founder works on your project.

  5. 06

    Owners’ mindset

    We’ve founded and run companies ourselves. When a founder faces accounting, corporate structure, a pitch deck or an investor report for the first time, we’ve been there — we’re not your accountants, but we’ll flag the mistakes before they cost you and challenge scope before we bill it.

  6. 07

    Clients stay 5+ years

    Several have been with us for over a decade. We build relationships, not projects.

Evgeny Smirnov, founder of 4xxi
05Founder

Evgeny Smirnov

PhD in astronomy, 20+ years in software, active researcher who publishes on LLM evaluation in peer-reviewed journals. He works on client projects himself. The habit he brings from research is simple: measure before you claim.

Also available as a fractional CTO or AI advisor.

06Our own products

Products we build for ourselves.

Each built from idea to release in a few days. We use the same pipeline for client prototypes.

Preduzmi

Accounting and tax compliance for Serbian lump-sum entrepreneurs. Invoices, KPO ledger and tax tracking.

preduzmi.com (opens in a new tab)

Marginalia

Expert-curated digests of scientific papers.

readmarginalia.com (opens in a new tab)
07Questions

What buyers ask us first.

Not here? Ask on a call, or write to [email protected].

What does it cost?

The AI opportunity audit is $5–10k, fixed, and takes 2 weeks. The prototype sprint is $20k, fixed, for 5 business days. Ongoing engineering is billed hourly; we share the rate on a call.

How quickly can you start?

Kickoff is typically within 3 business days of signing.

How do you keep AI from making things up?

The system answers only from your documents. Retrieval finds the passages and assigns the references, not the model. If nothing supports an answer, it says so instead of guessing — and we measure how often that happens before launch. Under the hood we combine deterministic techniques — retrieval scores, rules, assertions — with LLM-based verification to catch hallucinations and check every answer. The methods come from peer-reviewed research, including our own.

What if the audit isn’t useful?

The audit has a full money-back guarantee. If you sign a project with us within 60 days, the fee is credited in full toward it.

How does the prototype sprint work?

We agree a fixed scope up front, then build a working, deployable prototype on your documents and systems in 5 business days, for a fixed $20k.

Who does the work?

Senior engineers only, with no juniors and no handoffs. There is no sales layer: the engineers you meet on the first call build your project, and the founder works on it too.

Where does our data go?

Your data stays yours. EU/UK-based team, on-premises or private-cloud deployment, no training on your data.

Can you work with our existing team and codebase?

Yes, and we do it often. We join existing teams to add the expertise they’re missing — AI, architecture, delivery — and work in your codebase, your process and your tools.

Do you only build AI?

No. We have shipped 100+ products since 2008, across fintech, legal, education and consumer apps. AI is where we focus now.

Where are you based?

4xxi Software Ltd is registered in England and Wales. The team works across the UK and Europe.

08Contact

Let’s talk.

Tell us what you’re building or what isn’t working. You’ll hear back from an engineer, not a sales team.

Kickoff typically within 3 business days of signing.