Each one covers the problem, the decisions and the alternatives I rejected, the code that proves the claims, and what did not work.
LiveJan 2026 to present
A WhatsApp and Telegram agent whose calendar and search abilities live in a standalone MCP server, so the same server also plugs into Claude Desktop and Claude Code, with every message traced in Opik.
Open sourceJan 2026 to present
PDFs and photos in, clean Markdown out: tables as tables, maths as LaTeX, charts described in words.
ArchivedSept 2025 to Dec 2025
The first version: a LangGraph agent living inside WhatsApp with voice, images, calendar, web search and Qdrant memory, and the reasons its hand-wired tools had to go.
Workflow automation that takes repetitive work off people: agents and pipelines that run on events or schedules, move data between the tools a team already uses, and recover from failures on their own, with queues, retries and monitoring built in.
Designing and deploying production-grade ML pipelines, autonomous AI agents, and LLM-powered systems. From LangGraph orchestration to vector memory and multi-modal processing, built for scale and reliability.
Adapting foundation models to specialised domains using supervised fine-tuning, RLHF, and parameter-efficient techniques (LoRA / QLoRA). Resulting in sharper, task-specific models that outperform general-purpose prompting.
End-to-end product development from database design through REST APIs to polished React interfaces, delivered with performance, accessibility, and CI/CD built in from day one.
Long-form pieces on the parts of AI engineering that usually get skipped: protocols, retrieval, rate limits and the infrastructure underneath.
AI Engineering30 min read
Before MCP, every AI agent needed a custom integration for every tool it touched: the M by N problem that does not scale. MCP is the open standard that solves it, defining how AI applications discover tools, read resources and execute actions through a single...
AI / ML6 min read
Fine-tuning a model every time your data changes is costly, slow and forgets nothing on command. RAG stores knowledge in a vector database...
System Design6 min read
A single overloaded server is a single point of failure. Load balancing distributes traffic across a fleet, enabling horizontal scaling,...
Backend Engineering5 min read
An unprotected API endpoint is an open invitation to abuse, from scrapers and bots to cascading backend overload. Rate limiting enforces...
Production sites for studios, restaurants, cleaning, healthcare, governance and real estate clients across Uganda, Rwanda and the United States.
Virginia, United States
3 tools
6 tools
4 tools
2 tools
3 tools
3 tools
Kylie - A Multimodal WhatsApp AI Agent
We recently came across this project shared by Jonathan Mukhobe Sr. on LinkedIn and wanted to highlight it for the community.
Kylie is a fully functional WhatsApp AI agent built with a practical, end-to-end…
110+
GitHub stars on Kylie
Counted 18 September 2026
34
Forks
Counted 18 September 2026
10+
Production web and AI projects shipped
LinkedIn,
My original LinkedIn post about Kylie, the one that caught Qdrant's attention and led to the feature on X and LinkedIn.
Read the LinkedIn postThe website came out exactly as we imagined. Clean, fast, and professional. Our clients constantly compliment the look and feel, and bookings have noticeably gone up since launch.
Virginia, United States
Jonathan built us a site that truly represents who we are as a healthcare provider. The attention to accessibility and clarity made a real difference for our patients and families.
Home healthcare provider, Virginia
From the first call to the final handoff, the process was smooth and transparent. The platform is solid, our team loves using it, and we have had zero downtime since going live.
Technology and real estate, Kampala
Open to remote roles and contracts
If you are hiring for applied ML or agent engineering, or you have a system that needs to work rather than demo, get in touch. Pick whichever suits you; I reply within a day.
View on GitHub
