I build AI systems from Kabul, Afghanistan. Engineering here means fewer resources, slower infrastructure, and real users who do not tolerate latency as a design choice. It also means shipping things that work in imperfect conditions is the baseline expectation, not a differentiator.

My background is MERN — I can build full products fast and I know the entire stack well enough to make architecture decisions across it. The ceiling I kept hitting was clear: implementing features is one skill, designing systems with defined failure modes is another. I wanted the second one.

That pulled me into the actual mechanics of how LLM inference works: why a streaming pass cannot also be a tool-calling pass, how context window failure is not just a token budget problem, why recommendation engines become undebuggable when explainability is treated as a UI concern rather than a scoring concern. AICA is where I applied that thinking in a real product — two-phase Advisor pipeline, deterministic scoring with structured reasons, shared Zod contracts as a schema boundary between client and server.

The goal is AI Engineering Lead — not for the title but for the work. I want to own the architecture of systems that matter, name the failure modes before they happen, and make technical decisions that change what the product can actually guarantee to the user.