Every ML Model You Must Know, and When to Actually Use It
Which model for which problem, what each one is really doing under the hood, and where each breaks. Written for engineers shipping models and sitting interviews, not for a syllabus.
Hi, I’m Abdullah!
Sr. ML Engineer at Atlassian | Ex-Meta | PhD
RL & multi-Agentic AI for Confluence/Jira Search. On the Central AI team, I’m building and improving SMART Answer generation using RL-based fine-tuned LLMs and multi-Agent AI architecture. Previously at Meta on Ads Ranking, fine-tuning LLaMA 3 for large-scale suggestive ad generation.
I write about RecSys, agentic AI, RL fine-tuning, and what actually ships at scale.
Understand the Metrics Before You Build the Model
A practical map of ML metrics across four layers (offline, online, infra, business) and every model type from logistic regression to LLM agents, with the failure modes that make a good-looking number meaningless.
All About Optimization: From Gradient Descent to Muon
How models actually get trained: the optimizer family tree, why Adam won and where it loses, the schedule and warmup decisions that matter more than the optimizer choice, and the practical machinery (clipping, accumulation, checkpointing, sharded states) that shows up in real training runs.
Every Activation Function You Need to Know
Which activation to use where, why ReLU replaced sigmoid, why transformers moved to GELU and SwiGLU, and the failure modes (dead units, saturation, non-zero-centered outputs) interviewers ask you to diagnose.
Every Loss Function You Must Know as an ML Engineer
A scannable reference for ML interviews: which loss to use for which task, why it works, when it breaks, the PyTorch gotchas that cause real bugs, and how to answer the question every interviewer asks.
Python DSA Cheatsheet For Coding Interview
Python syntax and DSA patterns for coding interviews, condensed into a single reference: core syntax on page 1, the coding patterns worth memorizing on page 2.
The Biggest Gap in Agentic AI: Multi-Agent Evaluation
The missing layer between observability and benchmarks. Over the last year, Agentic AI has exploded. OpenAI has agents. Anthropic has agents. Google DeepMind has agents. Every startup suddenly has a multi-agent architecture diagram. And if you look closely, something interesting has happened. The industry solved observability. The industry largely solved benchmarks. Yet somehow, we still cannot answer a deceptively simple question: Was this agent actually good? Not “did it finish.” ...
The Missing Layer in Agentic AI: Why Evaluation Is the Next Enterprise Platform
Executive Summary Agentic AI is entering enterprise deployment faster than its evaluation infrastructure is maturing. Most teams can now observe traces and benchmark outcomes, but they still cannot reliably grade how agents behave in production across coordination quality, trajectory correctness, and safety compliance. That missing layer is becoming a strategic bottleneck for executive teams deciding where to place platform bets, set governance controls, and scale high-autonomy workflows with confidence. As of June 2026, the market has largely solved two layers: observability (OpenTelemetry GenAI conventions, AgentOps, OWASP AOS) and benchmark comparison (HAL, GAIA, SWE-bench). The unresolved layer sits between them: an open, framework-agnostic evaluation protocol that takes any OTel-compatible trace and scores agent behavior end-to-end. Without this layer, enterprises can measure activity and final outcomes, but still miss the process-level failures that drive hidden risk, cost overruns, and policy violations in real deployments. That gap is not only a research problem; it is now a platform opportunity with direct implications for deployment risk, governance, and competitive advantage. ...
The Evaluation of RecSys, Part 3: The Deep Learning Era (NCF, Wide & Deep, DeepFM, DIN, DLRM, AdaTT)
Part 3 of the series: how DNNs transformed RecSys from 2016 onward. NCF, Wide & Deep, DeepFM, DIN, DLRM, and AdaTT. Architectures, intuition, and where each shines.
The Evaluation of RecSys, Part 2: Factorization Machines and XGBoost
Part 2 of the series: how Factorization Machines generalized MF to arbitrary features, how XGBoost handled non-linear ranking, and the limitations that pushed the field toward deep neural networks.