SadafAmouzegar
About me.

I'm an AI/ML engineer, builder, and systems thinker focused on turning complex ideas into practical, scalable products.
Over the last decade I've worked across startups, aerospace, entertainment, and life sciences, building everything from production ML infrastructure and LLM-powered systems to tools for budgeting, analytics, and large-scale data intelligence. My work tends to live at the intersection of engineering, research, and product design: taking messy real-world problems and turning them into systems people can actually use.
But outside of the technical side, I'm someone deeply motivated by curiosity.
I've always been drawn to understanding how systems work, whether that's machine learning models, human behavior, language, art, music, storytelling, or the ways people make decisions under uncertainty. I love structure and organization, but I'm equally driven by creativity and emotional depth. A lot of my best work comes from balancing both sides.
I care deeply about building technology responsibly. Not just systems that are powerful, but systems that are thoughtful, reliable, and grounded in reality. I'm especially interested in how AI can augment human creativity and decision-making rather than replace it.
When I'm not building, I'm usually learning something new: studying languages, reading about machine learning and new breakthroughs in AI, writing poetry, making music, designing side projects, or disappearing into a topic until I understand it completely.
At the core of everything I do is the same instinct: to build things that are meaningful, elegant, and genuinely useful.
Where I've shipped.
Building production AI systems for medical device safety intelligence.
- Designed LLM workflows for analyzing FDA recall and adverse-event data
- Built evaluation and regression-testing pipelines against historical recall outcomes
- Developed backend services for retrieval, data processing, and model-assisted analysis
ML pipelines for LLM document analysis and classification, powering decision-support systems with reliability and interpretability built in.
- Risk-scoring workflows focused on safe, explainable model behavior
- Backend + product components across AI features (Python · Ruby on Rails)
- Evaluation frameworks to improve trust in AI outputs
Built an AI platform for film and TV pre-production, focused on budgeting, scheduling, and script analysis.
- Led product and technical direction from early prototype to production
- Built ML pipelines for script breakdowns, extraction, and production planning
- Deployed model-serving infrastructure on GCP supporting production users
Built NLP systems for brand analytics, market intelligence, and audience sentiment.
- Designed pipelines for trend detection and competitive analysis
- Built NLP models for extracting signals from public text data
- Led early product, data, and engineering direction
Built ML systems automating production workflows. NLP pipelines for script breakdowns and content generation; optimization for production planning.
- NLP engines for automated script analysis and story generation
- Predictive frameworks for production success metrics
- Generative-AI research foundational to a granted patent
Three-year ascent through avionics test software, internal tooling, and ML. Anomaly detection on engine and Dragon-vehicle test data.
- Anomaly detection across engine and stage test data; EMI testing on Dragon
- Models to prioritize and rank production issues by severity
- StarTrekker camera focus algorithm; calibration + video acquisition tooling
Undergraduate research in Babak Hassibi's lab on Hyperbolic vs. Euclidean Embedding of Real-World Networks. MDS against Crovella's online greedy embedding into the Poincaré disk.
- Implemented MDS + online greedy hyperbolic embedding in MATLAB
- Evaluated on preferential attachment, small-world, and Erdős–Rényi graphs
- Aerospace Corp. SURF Fellowship · co-mentored with Elizabeth Bodine-Baron
The toolkit, in production.
- LangGraph
- RAG pipelines
- Multi-agent workflows
- Agentic failure mode management
- Fine-tuning (LoRA, PEFT, RLHF)
- LangChain · LangSmith
- MCP
- Hugging Face
- PyTorch
- Evaluation & experimentation
- Deep learning
- NLP
- Model optimization
- MLflow
- TensorFlow
- scikit-learn · XGBoost
- Python
- GCP (Cloud Run, BigQuery, Pub/Sub)
- Vector databases (Pinecone, Qdrant, Weaviate)
- FastAPI
- CI/CD
- AWS
- Graph databases (Spanner, Neo4j)
- Docker
Press & selected work.
- 01DeadlineFounder coverageEnd Cue producers launch AI-powered production platform RivetAI2024↗
- 02Broadcast NowIndustry launchRivetAI launches AI production budgeting and scheduling platform2024↗
- 03At The Movies OnlineProduct featureRivetAI launches game-changing AI-powered production platform2024↗
- 04InverseGenerative-AI researchAI scriptwriting trained on Reddit creates an MST3K spoof2018↗
- 05The Next WebAI in entertainmentIBM Watson supercomputer crafts an AI film story2018↗
- 06Film University BabelsbergTalk · GermanyWorkshop · AI tools in film production2023↗
- 07Women of MENA in TechnologyProfile interviewPersian women in tech in the City of Angels, flying to new heights2018↗
- 08IMDbCreditsSelected filmography—↗
- 09Caltech · SURFAerospace Corp. Fellowship · Hassibi labHyperbolic vs. Euclidean Embedding of Real-World Networks2011↗