Ibrahim Denis Fofanah
Data Scientist & AI Researcher
Data scientist and AI researcher at Pace University. I coined Artificial Frictional Unemployment, and built the first machine learning model for crop yield prediction in Sierra Leone. Author of Understanding Agentic AI. I write about agentic systems and applied ML, with a bias toward what actually works, and who gets left out when it doesn't.
Stop Eyeballing Your Agent's Answers. A 100-Line Eval Harness Catches the Regressions You Miss.
Every agent builder hand-tests a few prompts and calls it evaluation. This tutorial builds a 100-line Python harness with a fixed case set, runs it against two versions of a mock agent, and catches a regression that eyeballing would have missed.
Ibrahim Denis Fofanah·Sep 27, 2026·7 min
Agent Reflection Does Not Beat a Retry. The Numbers From 16,946 Trials.
A 16,946-trial study finds Reflexion, Dynamic Cheatsheet, and ACE each beat plain retry in some settings and lose in others, while a scheduler called VEX2 came closest to a clean sweep. What practitioners should do about it.
Ibrahim Denis Fofanah·Sep 26, 2026·7 min
Your Retry Worked. That Is How You Got Two Rows.
Retries fix transient failures, but they can also duplicate payments, records, and side effects. Here is how idempotency keys, unique constraints, and atomic writes make retries safe.
Ibrahim Denis Fofanah·Sep 25, 2026·8 min
The Data Scientist Role Is Splitting Into Three Jobs. Here's Which One to Become.
The classic data-scientist bundle is unbundling into the analyst, the ML engineer, and the AI engineer. BLS still projects 34% growth, but the generalist entry seat is the one shrinking. Here is how to pick your one.
Ibrahim Denis Fofanah·Sep 25, 2026·7 min
What Are Embeddings? How Meaning Becomes Math
An embedding turns meaning into a location so that similarity becomes distance. That one idea is the engine under semantic search, recommendations, and the retrieval half of every RAG system, and it is cheap enough to run on your own data.
Ibrahim Denis Fofanah·Sep 25, 2026·7 min
3.1 Agent Workdays per Human Day Is Not 3.1× Productivity
OpenAI reports 3.1 agent-workdays of runtime for every human workday in its research organization. The number shows a major workflow shift, but runtime, useful output, and scientific progress are not the same metric.
Ibrahim Denis Fofanah·Sep 25, 2026·7 min
OpenAI Uses 3.1 Agent-Workdays for Every Human Day. The Bottleneck Moved.
OpenAI reports 3.1 agent-workdays for every human research day. That does not mean 3.1× productivity, it means the scarce work has shifted toward judgment.
Ibrahim Denis Fofanah·Sep 25, 2026·13 min
Data Storytelling Teaches You to Present a Finding. Nobody Teaches You to Find One.
Every guide teaches you to present a finding. Almost none teach you to find one. The four questions that break a story open, and a real Sierra Leone trial where every write-up missed it.
Ibrahim Denis Fofanah·Sep 25, 2026·7 min
OpenAI Got 6× Better CPU Efficiency With Rust. The Lesson Is Why They Waited.
OpenAI’s storage service once pushed more than 20 million requests per second through Python. A Rust rewrite later delivered 6× better CPU efficiency and 15× better memory efficiency. The useful engineering lesson is not “rewrite everything in Rust.” It is why OpenAI knowingly carried the slower system first.
Ibrahim Denis Fofanah·Sep 25, 2026·16 min
Quantizing Half the Attention Hurt. Quantizing Both Paths Barely Did.
Quantizing one attention path cost TabPFN-v3 as much as 31.8 Elo. Quantizing both cut the drop to roughly one point and enabled up to 1.7× faster inference. The result is a lesson about consistency, not just FP8.
Ibrahim Denis Fofanah·Sep 25, 2026·12 min
93% of the Hard Corporate Links Never Reach the Matcher
A new benchmark built from 6.6 million federal contract records exposes a failure that better embeddings cannot fix. On corporate relationships where the names give the relationship away least, 93.2% disappear before the matching model ever sees them.
Ibrahim Denis Fofanah·Sep 25, 2026·11 min
Your Model Works in a Notebook. That Is the Easy Part.
A notebook proves a model can work once. Production requires it to work continuously, on unseen data, within a budget, and to fail safely. The five questions that define the gap, and why deployment is the scarce skill.
Ibrahim Denis Fofanah·Sep 25, 2026·7 min
Want an author page like this?
We publish practitioners. Pitch an idea or republish a post from your own blog.