# Felipe Casadei

> Felipe Casadei leads mixed methods UX research at AWS, connecting behavioral science, AI evaluation, product strategy, and working prototypes.

Mixed Methods UX Research & AI Evaluation

Also published as Felipe Imbelissieri-Casadei.

## About

Research goes beyond job descriptions. I take a player-coach approach, combining technical skills with research experience.

I lead mixed methods UX research across four AWS life sciences products, including Amazon Bio Discovery, helping scientists accelerate drug discovery. My toolkit spans research prioritization, product incubation, prototyping and testing, fieldwork, experiments, surveys, supervised fine-tuning, and model evaluation.

Drawing on philosophy, neuroscience, and computational linguistics, I study knowledge, language, and human judgment. With engineering, applied science, and executive leadership, I shape AI behavior and examine its effects on consumer products and professional work.

[Currently accelerating Scientific learning at AWS](https://aws.amazon.com/biodiscovery/).

Based in Brooklyn, NY. My methods include A/B tests, large-scale surveys, telemetry, regression models, interviews, ethnography, usability testing, and longitudinal studies. My tools include Figma, Python, R, SQL, PyTorch, Hugging Face, and AWS Bedrock. Earlier work spans Alexa’s speech and language systems, prelaunch research for Alexa for Shopping, and risk and compliance research.

## Selected UX Work

[Full portfolio](https://felipecasadei.ai/portfolio): Password-protected interactive presentation.

- [Amazon Bio Discovery](https://felipecasadei.ai/research/amazon-bio-discovery): My research made scientific collaboration, learning, and inspectable AI central to Amazon Bio Discovery’s product direction. I led strategic and technical studies and built working prototypes that shaped shared workspaces, evidence controls, and architectural requirements, helping leadership commit resources to a broader scientific workbench. At the program level, the AWS–MSK collaboration subsequently [moved 100,000 antibody candidates from design to lab submission in weeks, compared with up to a year using traditional methods](https://www.aboutamazon.com/news/aws/aws-amazon-bio-discovery-ai-drug-research).
- [Alexa for Shopping](https://felipecasadei.ai/research/alexa-for-shopping): Product comparisons grew from thousands to a reported 10 million per month within 90 days of the team’s redesign. I led the international mixed methods research that shaped conversational behavior, guided discovery, and comparison features. Findings were reviewed by a CEO-led team and informed the data sources and architecture supporting the assistant. This contributed to a broader team effort, paving the way for an assistant later used by [over 350 million customers in the year to Q2 2026, as reported by Andy Jassy](https://www.aboutamazon.com/news/company-news/amazon-ceo-andy-jassy-stores-growth-ai-shopping-q2-2026-earnings).
- [Prime customer experience](https://felipecasadei.ai/research/prime-customer-experience): Prime teams gained a comparable view of customer experience across 22 countries, grounded in local membership value. I adapted the measurement framework and expanded quarterly surveys with conjoint analysis and contextual inquiry. The work improved the evidence used in business reviews, exposed local adoption barriers, including payment-method gaps, and helped teams prioritize product improvements. The program also secured an EU marketing pilot and later expanded to Prime sellers, extending its reach beyond my original remit.
- [Voice interaction at Meta](https://felipecasadei.ai/research/voice-interaction-at-meta): Established motorcycle couriers as a business segment for WhatsApp and shifted success measures from message volume toward interaction quality. Through fieldwork in India, I connected communication patterns with the cognitive demands of working in motion, informing voice interaction, accessibility, and product positioning across languages and professional contexts. The work contributed to a broader shift toward messaging as work infrastructure, in collaboration with product, design, and engineering.

## Prototypes, courses & academic research

- **[Petra](https://felipecasadei.ai/projects/petra)**: Petra is an AI harness for science, built to make model-generated claims traceable to their evidence. I’m developing a workbench that uses language models to extract claims and synthesize findings, with a graph connecting hypotheses to supporting and conflicting source passages. I separate model execution from scientific authority: a policy-controlled gateway routes bounded tasks, and people review proposed changes before they enter the trusted record. PostgreSQL holds that record; Neo4j supports bounded graph traversal so researchers can inspect why a claim is supported or challenged. For the mathematical foundations, my [graph theory course](https://graphtheory.felipecasadei.ai/) starts with notation and works up to knowledge graphs and GraphRAG. Status: In development.
- **Face Lab**: An active, on-device facial-analysis pipeline on Raspberry Pi, with a visual interface for experiments on prosody and expression.
- **[Graph Theory: From Zero to Diestel](https://graphtheory.felipecasadei.ai/)**: An independent sixteen-chapter teaching companion to Reinhard Diestel’s *Graph Theory*, 6th edition. I built visual explanations, worked examples, quizzes, and a Königsberg bridges game for readers without a background in mathematical notation. The course progresses from sets and proofs to knowledge graphs and GraphRAG; the AI chapters cite their own research sources.
- **[Amazon Resonance](https://felipecasadei.ai/projects/resonance)**: Researchers can turn a document of questions into a live survey, then analyze the responses in the same workspace as their other sources. Resonance uses an LLM to structure questions, add logic, estimate completion time, suggest improvements, and publish a shareable link. I designed and coded the workspace so analysis can continue while code, documents, and audio ingest in the background. Separate authoring and responding experiences share access and model contracts. Resonance is being tested within an AWS team of more than 100 people. Status: In testing at AWS.
- **[Felicity](https://felipecasadei.ai/projects/felicity)**: Felicity’s writing skills help AI assistants revise prose while preserving the author’s meaning. They draw on pragmatics, the study of meaning in context. [Grice’s cooperative principle](https://web.stanford.edu/class/psych205/papers/Grice-1975.pdf) informs checks for evidence, appropriate detail, relevance, and clarity; speech-act theory distinguishes stating a fact, making a request, and committing to an action. The Review and Write skills also examine common ground: what writer and reader can reasonably assume they share. Review flags consequential gaps and unintended implications; Write revises while preserving supported claims and voice. Both run in Codex and Claude Code. Status: Available on GitHub.
- **Prenatal Prosody Perception**: In design and experimentation. Recreates muffled speech heard before birth to test whether a small, brain-inspired network trained only on its rhythm and melody can distinguish languages and speech from non-speech.

## Petra demonstration details

The website’s “miniature investigation” panel shows a compact canvas preview of a completed graph and offers one authored walkthrough about AI reliability and GraphRAG’s reported results. Opening the panel starts an empty graph. Six selectable steps advance every four seconds, with back, play/pause, and forward controls. Brief captions distinguish a source-linked finding from an unsupported overclaim and leave the test of factual accuracy open. The graph fits its frame, with the source paper linked below. The demo uses Petra’s Cytoscape canvas library and the node shapes from its pictured focused graph, with a separate horizontal layout and playback. It makes no backend calls. The custom-question interface is deferred until its services are connected and tested.

## Education & Training

- University of Washington — MS Computational Linguistics (in progress; expected 2028)
- Columbia University — MS in Neuroscience
- Martinos Center, Harvard Medical School — fMRI Research Fellow
- University of São Paulo — BA Philosophy

## Past Industry

- Amazon Web Services — Mixed Methods UX Researcher, Pharma & Life Sciences AI (2025–present)
- MITRE — Principal AI Engineer, generative AI (2025)
- U.S. Dept. of State — Foreign Service Officer (2024–2025)
- Amazon — Mixed Methods UX Researcher, Alexa for Shopping, Prime & Voice AI (2020–2024)
- letgo — UX Research Manager (2020)
- Meta — UX Research & Insights Program Manager, International Growth (2018–2020)
- Mastercard Advisors — Senior UX Researcher (2017–2018)
- Google — Research Program Manager / UX Researcher (2014–2017)
- Google — Trust & Safety Investigator (2012–2014)
- YouTube (contract) — Trust & Safety Coordinator (2011–2012)

## Writing

Essays on Medium: https://medium.com/@felipe_cas

## Podcast

Book and paper summaries on dratsi-pod, from philosophy and history to neuroscience and mathematics.
Spotify: https://open.spotify.com/show/7cH9c1aV3hg2jVIKEqUIWB

## Open To

Staff UX Research and Design Technology roles shaping AI products, plus research collaborations, advising, and speaking.

## Links

- Site: https://felipecasadei.ai
- LinkedIn: https://www.linkedin.com/in/fecasadei
- GitHub: https://github.com/beethovensaves
- Medium: https://medium.com/@felipe_cas
- Spotify: https://open.spotify.com/show/7cH9c1aV3hg2jVIKEqUIWB
- Contact: felipe@dratsi.com