Agent Handoff Kit
A Python and SQLite recovery prototype that resumes interrupted local work from a durable checkpoint. Its guarantees apply to the local database, not external APIs.
Letlhogonolo Fanampe
AI Assistants • RAG Systems • Workflow Automation
I build and deploy AI applications that connect business knowledge, data and workflows. My work spans retrieval-augmented generation (RAG), SQL assistants, API integrations, document processing and AI quality evaluation.
Pretoria, South Africa · Available immediately for AI Solutions Engineer, AI Engineer and LLM application engineering opportunities, including remote and hybrid roles.
DJ Papzin · Builder · DJ since 2012
Selected work
A Python and SQLite recovery prototype that resumes interrupted local work from a durable checkpoint. Its guarantees apply to the local database, not external APIs.
Connects a FastAPI dashboard and Telegram commands to a SQLite run log for goals, approvals, history and model routing. Demo mode works without provider credentials; live integrations require configuration.
A team project exploring news classification with Llama 3 and Together AI. My role: AI Specialist. The repository documents model fine-tuning on WELFake and a Python/Flask classification API.
Who
I build and deploy AI applications that connect business knowledge, data and workflows. My work spans retrieval-augmented generation (RAG), SQL assistants, API integrations, document processing and AI quality evaluation.
At Kwantu, I developed RAG and SQL chatbot solutions using Python, FastAPI, LangChain and vector databases. My work included multilingual information retrieval, WhatsApp and Telegram integrations, OCR pipelines, and deployment on Linux with systemd and Nginx.
Through work with Outlier and Translated, I have also evaluated model responses, designed prompts and contributed human feedback for AI training. That experience informs how I test applications for accuracy, relevance and failure cases.
Before moving into software, I spent eight years in AfriSam’s cement laboratory. That work taught me to check data carefully, investigate unexpected results and make processes reproducible. I started learning Python in 2022 and developed my skills through projects and AI hackathons.
When I’m not building, I’m DJing. I’ve been at it since 2012 and co-founded Papzin & Crew in 2016.
Available immediately for AI Solutions Engineer, AI Engineer and LLM application engineering opportunities, including remote and hybrid roles.
Stack
Proof
One-Day Challenge — built VisualPro, a WebGPU backend with 3D data visualization using Three.js and Node.js.
Built Vocal Thread — YouTube comment-to-speech with sentiment analysis using ElevenLabs API and Google Gemini.
One-click solution launcher integrating GitHub, Gitpod, and Claude AI (Anthropic API).
Learning by building in AI hackathons, with certificates for events including Llama 3, Falcon, Autonomous Agents, and AI Agents with MindsDB.
View participation & certificates ↗Career
Co-founded Papzin & Crew with Gabriel Matshabe in 2016. The platform features mega-mixes, Cruize Friday mixes, online radio and custom mix requests.
Foundation
Professional development: LangChain and Vector Databases in Production · Python · System Administration.
Work
A Python and SQLite recovery prototype that resumes interrupted local work from a durable checkpoint. Its guarantees apply to the local database, not external APIs.
The demo interrupts one worker and lets a replacement resume, keeping one durable receipt and verifying unchanged replay. No model API keys or pip dependencies are needed for the local demo.
Connects a FastAPI dashboard and Telegram commands to a SQLite run log for goals, approvals, history and model routing. Demo mode works without provider credentials; live integrations require configuration.
Includes a recorded demo, Docker setup and optional integrations with model providers.
A team project exploring news classification with Llama 3 and Together AI. My role: AI Specialist. The repository documents model fine-tuning on WELFake and a Python/Flask classification API.
The interface displays a classification and confidence score; its output should be checked against reliable sources.
A Streamlit application that analyses YouTube comments and turns them into spoken dialogues using ElevenLabs, YouTube Data API and Google Gemini.
Includes sentiment analysis and a Chat with Comments feature. Built for the ElevenLabs AI Audio Challenge.
A Django-based computer-vision tool that extracts messages, timestamps and emojis from chat screenshots using YOLO and PaddleOCR.
What it does: Turns an uploaded screenshot into structured message data through an extraction API. The repository includes sample inputs and outputs.
DJ platform connecting Mega-Mixers with music lovers. Freemium streaming, Cruize Friday mixes, custom mix requests, and 24/7 online radio. Co-founded in 2016.
What it does: Connects Mega-Mixers with listeners through streaming and custom mix requests.
↗ GitHubAI-powered task management with FastAPI, PostgreSQL, and state-machine enforcement for engineering workflows. Powers PapzinAI operations.
What it does: Enforces task transitions for engineering workflows.
Enforces task transitions for engineering workflows. FastAPI exposes the task workflow while PostgreSQL stores its state. State-machine checks govern transitions; the project supports PapzinAI operations.
Multi-agent AI system orchestrating autonomous coding agents — Hermes, OpenClaw, and custom workflows for automated development.
What it does: Coordinates coding agents and automated development workflows.
Coordinates coding agents and automated development workflows. Hermes, OpenClaw and custom workflows form the orchestration described here. The focus is connecting coding agents with repeatable development operations.
WebGPU backend with 3D data visualization. API endpoints with Node.js, SQLite storage, Three.js rendering. WebGPU Hackathon winner.
Outcome: First place at the WebGPU Hackathon.
Smart recruitment assistant handling candidate screening, interview scheduling, and feedback collection. NLP-powered HR automation.
What it does: Brings screening, interview scheduling and feedback into one workflow.
Brings screening, interview scheduling and feedback into one workflow. NLP supports candidate screening, followed by interview scheduling and feedback collection. Contact me for implementation details or to discuss a similar workflow.
Drop a track or hit Demo to see the visualizer react
Listen
DJing since 2012. Co-founded Papzin & Crew in 2016 because Spotify playlists are boring and Mega-Mixers deserve a platform. Cruize Friday mixes, custom requests, 24/7 streaming — real music by real people.
Now I'm also experimenting with AI-generated music through Suno. Because why not? If the AI can make something that moves people, I'm in.
Connect
Available immediately for AI Solutions Engineer, AI Engineer and LLM application engineering opportunities, including remote and hybrid roles.