Los Angeles · Netflix · University of Illinois Urbana-Champaign
Hi, I'm Mariam,
AI Engineer.
What I work on
Four threads I keep pulling on — at work, at school, on the side.
Applied AI
GenAI applications I build at Netflix and on the side — agents, retrieval systems, and tooling that gets used.
- →AI agents
- →RAG pipelines
- →Evals & benchmarks
- →Orchestration
Research
Areas I'm digging into through my Master's and personal reading.
- →Generative AI
- →Deep Learning for Healthcare
- →LLM Evals
Analytics Engineering
How I spend my days at Netflix: turning messy data into dashboards and decisions leaders can act on.
- →Data modeling
- →Dashboards
- →Pipelines
- →Business metrics
Writing & Talks
Notes, writeups, and the occasional presentation on what I'm learning.
- →Technical essays
- →Project writeups
- →Notes
- →Talks
About
I'm a Senior Analytics Engineer at Netflix on the Growth Data Science & Engineering team, where I build GenAI applications, internal tooling, and dashboards that help leaders make better business decisions. Before Netflix, I was a Full Stack Developer at a Y Combinator startup — a role I came into through UCLA's Full Stack Web Development certificate program.
On the side, I'm finishing a Master's in Computer Science at the University of Illinois Urbana-Champaign — graduating after Fall 2026. Most of my coursework and personal projects sit at the intersection of generative AI, data systems, and applied research.
Selected work
Things I've built, contributed to, or created recently.
Predicting Chest X-Rays from EHR
Team re-implementation of the EHRXDiff framework — Latent Diffusion + U-Net — for predicting a patient's future chest X-rays from their prior imaging and electronic health records. Replaced external API embeddings with locally hosted BERT encoders for privacy, and added RAG to better capture rapid ICU signal changes. Contributed back via a PyHealth pull request.
Hate Speech Detection Survey
A comprehensive literature review of NLP approaches to detecting hate speech in online text — from early rule-based systems like Spertus's Smokey, through classical ML classifiers, to current deep learning methods. Covers historical and legislative context, key benchmark datasets, and where the field is heading.
Predicting Annotation SLAs with MLR
Multiple Linear Regression model predicting service-level agreements for data labeling jobs, trained on 50k+ rows of annotator-team performance data. Engineered features from timestamps, addressed multicollinearity via PCA, and evaluated against time-series alternatives.
Project Four
TODO: load previous projects.
What are you working on right now?+
At Netflix, GenAI applications and internal tooling for the Growth Data Science & Engineering and our GenAI Task Force. On the school side, my Master's coursework at the University of Illinois Urbana-Champaign — includes applied ML and data science. On the personal side, small projects exploring agentic systems while cantoring every weekend at St. Brendan in Los Angeles
When do you graduate?+
After Fall 2026. The Master's in Computer Science is at the University of Illinois Urbana-Champaign, and I've been doing it part-time alongside my work at Netflix.
Where can I see more of your work?+
Code lives on GitHub, research is here in this site. My work at Netflix is internal.
What's your stack?+
Python, Druid and Spark for analytics and data; Posit Connect, TypeScript / React / Next.js for the application side. GenAI usual suspects — LLM APIs, vector stores, eval frameworks. Always evaluating what's new.
Are you available for freelance or side projects?+
Yes — if you've got a well-thought-out idea, I'd love to hear it. Tell me what you're trying to solve, what you've already tried, and where you think I'd add value.
Where are you based?+
Los Angeles. Netflix is local; my Master's at the University of Illinois Urbana-Champaign is remote.
Writing & notes
Things I'm thinking about.
Agent Memory
How to handle context window across long-horizon tasks — what to keep, what to forget, and where state lives between turns. Reading on memory hierarchies, retrieval-augmented memory, and the different design choices to keep high quality agent output over time.
Agent Core Runtime
AI agent architecture — tool routing, the inner loop, prompt assembly, context management, and how harnesses keep agents grounded across long sessions. Especially interested in how memory, tools, and planning compose into something that holds up in production.