Real AI & Data Systems,
Built Under African Constraints.
Most technical content assumes stable power, cheap GPUs, and fast internet. I document what actually works in production — systems engineered in Kenya, tested against the constraints most tutorials never mention.
Read by engineers building real systems.
Trending
Data Engineering
Pipelines & ETL Systems
Trending Now
Apache Data Lakehouse Weekly: April 16–22, 2026
Dev.to
How I scrape and de-dupe Meta ads for 1000 brands
Dev.to
ClickHouse Native JSON Support in 2026: A PR-by-PR Analysis
Dev.to
# Building a Streaming Session Analytics Pipeline with Kafka, Postgres, and dbt
Dev.to
The Announcement Everyone Slept On at Google Cloud Next '26: The Cross-Cloud Lakehouse
Dev.to
Artificial Intelligence
LLMs & Deep Learning
Trending Now
Show HN: Pantheon-CLI – Open-Source Python Claude Code and Smart Notebook
Hacker News
Congrats to the Notion MCP Challenge Winners!
Dev.to
If AI Existed in 2011, Would We Still Have the Modern Web?
Dev.to
Defluffer - reduce token usage 📉 by 45% using this one simple trick! [Earthday challenge]
Dev.to
It Might Already Be Too Late to Fix This
Dev.to
Machine Learning Systems
Model Training & Deployment
Trending Now
Building a Fully Automated Horse Racing AI Prediction Pipeline with Flutter + Supabase
Dev.to
Fine-Tuning LLMs for Legal Tech: Nebius AI Cloud vs Nebius Token Factory — A Developer's Honest Comparison
Dev.to
Introduction to Machine Learning
Dev.to
Why AI Systems Don’t Fail — They Drift
Dev.to
Understanding Transformers Part 8: Shared Weights in Self-Attention
Dev.to
AI Automation
Agents, MLOps & Workflows
Trending Now
I Thought Fine-Tuning Needed an ML Team. I Was Wrong.
Dev.to
How I built an AI platform in a country where the Western ones aren’t for sale
Dev.to
GenAIOps on AWS: Production Hardening & Advanced Patterns - Part 4
Dev.to
Why the Line Between Data Engineer and ML Engineer Is Disappearing, And Why That's Your Cue to Cross It
Dev.to
Production GPU Training is 34% Slower. Show Me Why
Dev.to
From the Field
Systems analysis, architecture deep dives, production postmortems, and applied research. Written from experience, not speculation.
Building a RAG System That Survives Production
Beyond the demo: chunking strategy, hybrid retrieval, latency budgets, and why your embedding model matters more than your vector DB.
Hallucination as an Engineering Problem: H-Score
Most teams treat hallucination as a model quality issue. I built a metric to measure it as a system property.
When RL Meets RAG: Systems That Know What They Don't Know
Combining reinforcement learning with retrieval to build enterprise AI that adapts without hallucinating.
How M-Pesa Processes 10 Million Transactions a Day
Inside the queuing, validation, and settlement layers keeping East Africa's most critical financial rail running under load.
Inside a Real-Time Fraud Detection System
Architecture breakdown of a streaming fraud detection system: feature engineering, model serving, and the false positive problem.
Why Our Vector Search Was Slow: A Postmortem
HNSW index misconfiguration, chunk boundary issues, and the query pattern we never tested. A debugging session, documented.
Designing for Intermittent Connectivity
What most guides skip: building data pipelines that handle 4-hour connectivity gaps without data loss or downstream inconsistency.
Building AI for Environments That Don't Fit the Defaults
What changes when you're building AI systems for hard cost ceilings, unreliable APIs, and users whose defaults differ from Silicon Valley's.
Three Ways Our ML Model Broke in Deployment
Data drift, silent failures, and a misconfigured feature store. A postmortem on what we missed in testing.
The Hidden Cost of Switching Embedding Models
We changed our embedding model mid-project. Here's what broke, what we didn't anticipate, and what re-indexing at scale actually costs.
2 published · 10 in progress
Ongoing
Signature Series
A 3-part production series. Read in order. Each part builds on the last.
PART 01
Signature Series
How It Actually Runs
Real systems, explained without the marketing diagram.
Every system that runs at scale has a gap between the architecture diagram and what's actually deployed. This series tears down real systems — payments infrastructure, data platforms, ML pipelines — and explains exactly how they work in production, constraints included.
In this series
How M-Pesa Handles 10 Million Transactions a Day
Inside a Real-Time Fraud Detection Pipeline
How Recommendation Engines Work at Low Data Volume
PART 02
Signature Series
If I Built This Today
Retrospective system design. What experience actually changes.
Looking back at systems I actually built in production. Not a highlight reel — the real decisions I made, why I made them given the constraints at the time, and what I would do completely differently with what I know now.
In this series
The Data Pipeline I Regret Building
Our First ML System: Every Wrong Assumption, Documented
Rebuilding a Feature Store After It Failed at Scale
PART 03
Signature Series
What Breaks in Production
Honest postmortems. No blame, no hand-waving.
Production failures documented in full: what the symptoms were, what we initially diagnosed, what was actually wrong, and what changed so it would not happen again. Written as engineering records, not incident theater.
In this series
The Silent Data Loss We Found Three Months Late
When Our Model Passed All Tests and Failed in Production
The Latency Regression We Shipped to 100k Users
Start with Part 01 and read in sequence for the full picture.
Applied System Design
Research
A Reliability-First Framework Combining Adaptive AI with Hallucination Safeguards
Autonomous AI agents in enterprise create a specific tension: the system needs to adapt to new information, but it also needs to produce outputs you can stake a business decision on. I designed a framework that addresses both — combining Reinforcement Learning, Retrieval-Augmented Generation, and uncertainty quantification. The core contribution is the H-Score, a metric I built to measure hallucination risk as a system property across IT operations, healthcare, and logistics environments.
H-Score
A metric for measuring hallucination risk in enterprise AI systems
CWAS
Confidence-Weighted Action Selection: RL that knows when to be cautious
REAC
A benchmark corpus for testing enterprise AI agents against real workflows
Experiments · Prototypes · Real Implementations
System Lab
Real systems built under real constraints. Each entry documents the problem, the design decisions, and what actually happened — including what failed. Not a portfolio. An engineering record.
About the Author
Sam Odongo
Nairobi, KenyaSenior Data Engineer · Applied Data Scientist
I build AI and data systems where most architectures fail. Low bandwidth, unstable infrastructure, real users, real constraints.
Most technical content assumes stable power, cheap GPUs, and fast internet. I work in Nairobi, Kenya, where none of those are guaranteed. That gap between theory and production is where I live and where I write from.
I hold an MSc in Artificial Intelligence and have spent years building production data systems across fraud detection, real-time streaming, RAG pipelines, and MLOps infrastructure. Every system I document here has been tested against constraints most tutorials never mention.
If you are building systems that have to work in the real world, not just on a benchmark, this is where you should be.
Join engineers learning how to build AI systems that don't break in production.
Real systems. Real constraints. Written from Nairobi, tested against the infrastructure most tutorials never mention. No spam. No theory. Unsubscribe anytime.
Written by a practitioner, not a marketer.