Johannesburg, South Africa · SAST (UTC+2)

Katlego Mokoto

AI & ML Engineer / Full-Stack Developer

I build AI systems that do the slow work and stop for a human yes: RAG assistants, career pipelines, research desks and on-device tools on .NET/C#, React and Python. Open them and test them here.

AI & ML Engineer
Full-stack developer
AI systems architect
Product engineer

Availability: Available immediately
Engagement: On-site, hybrid, or remote

Johannesburg, Pretoria, Cape Town, Durban, and remote teams

See the four live systems View CV

LinkedIn Contact 065 842 4274

Request a code walkthrough

Portrait of Katlego Mokoto — AvailableMens Mercatus

Selected work

Four systems worth an interview

Open the case studies

01 · Career pipeline

Human approval required

JobNet Application

A daily career pipeline that turns public job posts into a ranked shortlist and tailored drafts, and sends a letter only when the posting prints an employer address.

Role: Designed and operated the pipeline
Stack: Agents · Gmail · macOS automation · CV tailoring
Result: 20:00 SAST scrape, sentence ranker, one role per company. A note is sent only when the posting prints an employer address, that mailbox verifies, and the letter cites the role. Otherwise the send is held. The previous run is reported at 15:00 SAST.

Read case study Open JobNet Application

02 · Multimodal assistant

Live, with a labelled fallback

Mercatus Assist & THINK

A C# / ASP.NET Core service and a React workspace that call Gemini from the server, keep the API key off the browser, and say so when the live path is down.

Role: Built the API, the proxy, and the workspace
Stack: C# · ASP.NET Core 8 · React 19 · TypeScript · Gemini · nginx · systemd
Result: Provider-swappable chat completions, exponential backoff on 429 and 5xx, per-IP caps, and a labelled demo mode when the robot is unhealthy.

Read case study Open Mercatus Assist

03 · Research system

Paper research desk

Forex Mercatus Codex

A hosted Python terminal for charts, multi-market views, and a session-scoped paper bot — built so it cannot be mistaken for a brokerage.

Role: Built, hosted, and bounded the desk
Stack: Python · Highcharts · yfinance · Ubuntu VPS
Result: About 87 MB on disk, quotes refreshing near every 60 seconds, up to 24 sessions with a 2-hour idle TTL, and a public health JSON.

Read case study Open the paper desk

04 · Browser studio

Runs in the browser

SoundX AI

An AI drum station and studio with 20 on-device AI tools, a plugin rack, Afro sample packs and 22 themes. Audio stays on the machine. No sign-in.

Role: Built the studio, the on-device AI tools and the Producer Agent
Stack: React · Web Audio · TypeScript · on-device ML
Result: A brief becomes a full beat in one click. Tempo and key detection, hum-to-MIDI, clash-fixing EQ and Mix QA all run in the browser.

Read case study Open SoundX AI

How the systems are built

Agentic workflows, with a person on the irreversible step

AI-assisted means a model returns a bounded output: a draft, a summary, a setting. Agentic means the system can pick an approved tool, keep state, and propose the next step. It does not mean the system sends mail, places orders, or hides a failure.

01
Intent
What the person asked for
02
Context
Only the data that task needs
03
Router
Model, tool, or fallback
04
Tools
Approved APIs and files
05
Human gate
Stop before irreversible acts
06
Record
Status, latency, source

About

Who you would be hiring

I want a team that ships applied AI with clear ownership: model calls behind a service, a human decision before anything irreversible, and production habits — retries, rate limits, health checks — on the first public version.

Self-taught. 4 years’ experience. Available immediately.

Read the résumé

Mar 2026 – present
AI systems architect and founder
Project Antigravity
Ongoing
Full-stack, .NET, and AI systems
Mens Mercatus
Before the portfolio
Operations and optimization
Earlier systems work