AI that survives an audit
Agents, LLM platforms and retrieval systems built the way a bank would want them: evaluated against a metric, costed per call, logged end to end and explainable to whoever asks.
- LLM platforms
- Agents
- RAG
- Evaluation harnesses
Tech & AI · South Africa
Four dogs.
One hell of a pack.
HondSlim is four senior engineers with ten-plus years each in finance — the industry where it has to be right, fast, auditable and live by Monday. We build AI, data platforms and products that go to production and stay there.
Shipped inBankingPaymentsInsuranceCapital marketsCredit & lending+17 more
What we do
Four people. Absurd range.
Small enough that you talk to the people writing the code. Senior enough that there is nothing to hand off to. We take the whole thing: the AI, the data underneath it, the product around it, and the regulation all three live inside.
Real-time data platforms
Streaming, lakehouses and pipelines that do not need babysitting. Millions of records a minute, reconciled, queryable and boring in the best possible way.
- Kafka
- Flink
- Iceberg
- Postgres
Product engineering
Whole products, not proofs of concept. Web, mobile, API and the deployment path underneath, from the first wireframe to a thing paying customers log into.
- SvelteKit / React
- Go
- AWS & Cloudflare
- CI/CD
Fintech & compliance
Payments, credit decisioning, fraud and the regulation around all three. We have built inside the rules often enough to know where they actually bind.
- Payments
- Credit & risk
- Fraud
- POPIA / GDPR
Live, in your browser
This is the whole job, in one picture.
Grey dots are your data: scattered, drifting, no shape to it. The four coloured agents are us. Nothing here is a loop — each dog picks the point furthest out, swings around behind it and drives it in. Points that reach the pen snap into a lattice and go from noise to structure. Shove the herd with your cursor, or click to scatter the lot and watch the pack start again.
A simulation of four agents herding scattered data points into an ordered lattice. It needs JavaScript and a motion preference that allows animation.
The pack
Named after the colours. Answer to anything.
Four of us, every one named for an Afrikaans colour, every one with ten-plus years in finance behind them. No bench, no juniors wearing a senior's rate card, no account manager between you and the person writing the code.
Goue Hond
Chief Executive Officer
The one who decides
Runs the pack, reads the room and closes the deal. Fifteen years turning "that is impossible in this regulatory environment" into a signed statement of work and a shipped system.
- Strategy
- Delivery
- Regulated industries
- Saying no
Rooi Hond
Chief Technology Officer
The one who builds it
Architecture, infrastructure and the 3am pager. Has personally carried platforms doing millions of transactions a day, and has the strong opinions to prove it.
- Distributed systems
- Cloud & Kubernetes
- Go / Rust / Python
- Security
Groen Hond
Head of AI
The clever one
Models, agents and retrieval that hold up under audit. Takes AI past the demo: evaluated, costed, monitored and defensible in front of a risk committee.
- LLM & agents
- RAG & retrieval
- Computer vision
- MLOps
Pers Hond
Head of Marketing
The loud one
Makes the noise that gets you noticed. Positioning, brand and go-to-market for products that are genuinely good and that nobody has heard of yet.
- Positioning
- Brand
- Go-to-market
- Growth
Hond is Afrikaans for dog and slim for clever. Pers happens to mean both purple and the press, which is either a coincidence or the best hiring decision we ever made.
Our own product, in production
DigiVacc
Digital proof of animal vaccination for South Africa: a vaccination card every owner actually has on them, and a legally credible certification layer where an SAVC-registered vet e-signs the official certificate. One platform, built end to end by this pack.
- Ownership transfers that carry an animal’s full history with it
- Practice software that issues real records before an owner has even signed up
- Certificates that stand up to the rules they are written under
Where we have been
Finance first. Then everywhere else.
Ten years each inside banking, payments and credit — the hardest place to ship software and the best possible training for everywhere that is easier. That bar came with us into every industry on this list.
- Banking
- Payments
- Insurance
- Capital markets
- Credit & lending
- Fraud & risk
- Wealth management
- Crypto & digital assets
- Healthcare
- Veterinary
- AgriTech
- Logistics
- Retail & e-commerce
- Telecoms
- Energy & utilities
- Mining
- Manufacturing
- IoT & industrial
- Media
- EdTech
- PropTech
- Government
Not on the list? It is a list, not a limit. The constraint was never the industry — it was always the data.
How it goes
Scent. Herd. Drive. Hand over.
The same four moves every time, because they work. You will know within two weeks whether this is going anywhere, and you will own the result outright at the end of it.
- 01
Scent
Find the actual problem
A short, paid discovery. Your data, your constraints, your regulator. You get a written plan and a straight answer, whether or not we are the ones to build it.
- 02
Herd
Get the mess into a shape
The scattered thing becomes a scoped thing: a metric, a baseline, an architecture and a sequence that puts something working in front of you in weeks.
- 03
Drive
Ship in short bursts
Two-week iterations in your environment, behind your CI, against the metric we agreed. Every one of them ends with software in production, not a slide.
- 04
Hand over
Leave you able to run it
Runbooks, documentation and pairing until your team owns it outright. A good sheepdog gets the flock home and then sits down.
Track record
Numbers, not adjectives.
Every figure below came off something we built, shipped and then had to keep running at 3am. Ask about any of them.
- 40+ yrs
- Combined experience
- four seniors, no juniors billed as seniors
- 10+ yrs
- Each, in finance
- banking, payments, credit and risk, inside the regulation
- 7,000,000+
- Transactions a day
- scored by decisioning systems we built and ran
- 1,000,000+
- Records a minute
- through near real-time lakehouse and streaming platforms
Sic us on it
Tell us what is not working.
A platform nobody will touch. An AI pilot that will not graduate. Data in nine places and none of them agreeing. A product that needs building from nothing, this quarter. Describe it in a paragraph and you will get a straight answer about whether we are the right pack — including when we are not.