Caleb Andersen Applied AI Engineer · Durban ZA

Currently · Solutions Architect at ColorGen · Founder of Archeta

I ship AI products, then defend the decisions behind them.

Backend and applied-AI engineer from Durban. I build delivery systems, retrieval pipelines, and industrial automation that have to work in production. Explaining why a system is built the way it is, I'd argue, is half the job. Founder of Archeta, accepted into Microsoft for Startups; creator of the open-source RagCap.

Selected work

04 case studies
01
ARC · Product

Archeta

An AI delivery-confidence engine for small dev teams. Runs async standups, reads GitHub and Slack into one authoritative delivery state, flags risk early, and recommends the single next move. It deliberately stays out of prioritization and code.

MSfor Startups · Tier 2
Read case
02
RAG · Open source

RagCap

Portable RAG as a single file. Bundle sources, chunks, and embeddings into one .ragcap SQLite capsule you can build, search, ask, serve, and export. It's offline, reproducible, and CLI-first, with local ONNX or OpenAI/Azure providers.

1 filebuild · ask · serve
Read case
03
VIS · Hackathon

Evidence Review

A multi-modal claim-verification agent that checks whether submitted photos actually support a damage claim across cars, laptops, and packages. One reasoning agent with a real tool loop inspects each image, weighs it against the claim and the user's history, and decides: supported, contradicted, or not enough information.

5thof 1,773 finalists
Read case
04
TRI · Hackathon

Orchestrate Triage

A deterministic support-triage pipeline that classifies and routes requests across many companies without hallucinating policy. It runs on lexical retrieval, rule-based routing, and a modular, testable pipeline instead of a freeform LLM.

70thof 1,349 finalists
Read case

How I work

principles
/01

Deterministic over clever

When a system touches policy, money, or production, I reach for rules and retrieval before generation. Predictable beats impressive.

/02

Boundaries are a feature

Archeta refuses to prioritize or write code on purpose. Knowing what a system should not do is how it earns trust.

/03

Fail fast, fail loud

Schema-drift checks on startup, closed-loop verification, no silent migrations. I'd rather a service refuse to boot than corrupt state quietly.

/04

Ship, then defend

Anyone can build a demo. I write the tradeoffs down and stand behind them. That's what the case studies on this page are for.

"Building software is only half the challenge. Understanding and defending your technical decisions matters just as much." — what placing 70th, then 5th, taught me

Contact

open to AI / backend work

Building something that has to actually work? Let's talk →