Amin Mirlohi · PhD, Computer Science
Independent builder and AI engineer working at the intersection of retrieval, multi-agent systems, and the discipline of making models reliable. I build real products and write about the engineering, not the hype.
What I work on
Four threads run through everything I build, from research prototypes to production systems. Each leads with the same question: how do we know this actually works?
Designing retrieval-augmented systems that ground model output in real sources: hybrid retrieval (lexical + dense), semantic chunking, reranking, and citation-grounded answers.
Building agent systems that plan, call tools, and manage state reliably, with deterministic guards, retries, and explicit boundaries instead of hopeful prompting.
Treating evaluation as engineering: golden datasets, faithfulness and groundedness metrics, regression gates, and adversarial testing so quality is measured, not assumed.
Bringing PhD-level research rigor to production: information retrieval, network analysis, and adversarial/applied machine learning, translated into systems that ship.
Latest writing
AI can build a component from a screenshot. It still can't tell you whether it matched. The fix is the same one that applies to every agentic loop: separate the measurement from the judgment.
Anthropic's new Claude Tag turns @Claude into a persistent, shared teammate inside Slack. Here's what it actually does, how it's governed, and why the form factor matters more than the feature list.
The unit of agentic work has moved from the prompt to the loop. A working definition, an anatomy of what a loop is made of, and the one principle that separates loops you can trust from agents that agree with themselves.
If you care about retrieval, agents, or evaluation done properly, the writing is the best place to start. For anything else, the door is open.