yasaman@kot:~$ whoami
Research Assistant at MIT FutureTech and WashU School of Medicine. I build LLM pipelines that turn unstructured documents into measurements, and agents that remember.
yasaman@kot:~$ whoami --verbose
I'm a Research Assistant at MIT FutureTech, a joint group of MIT CSAIL and MIT Sloan, where I build LLM pipelines that measure firm-level AI adoption from large corpora of SEC 10-K filings and job postings, structured with the APQC Process Classification Framework.
At WashU School of Medicine I build a RAG pipeline that reads clinical trial documents for radiotherapy constraints. Earlier, I taught an LLM agent to live in Minecraft. Before that: PHP, PL/SQL and Oracle RAC. I was moving data through pipes long before the pipes learned to think.
What keeps me curious: scalable LLM pipelines, agentic systems, and what the architecture of minds can teach the architecture of AI.
yasaman@kot:~$ whoami --temperature 1.2
A former Oracle DBA who now teaches language models to read documents that were never meant to be read: corporate filings, clinical protocols, game logs. Pretrained on databases, fine-tuned on research, currently deployed at MIT FutureTech.
Known behaviors: turns every problem into a pipeline; keeps a Minecraft agent that once wandered off mid-experiment; believes memory is the most under-rated part of an agent; reads poetry between eval runs.
Context window: long. Patience for unlabeled data: shorter.