Wesley Wei Qian
Experience
VP, Engineering & Research (Founding Team) | Osmo
Sep 2022 - Present
- I started as the first research engineer, grew into the first engineering manager, and now lead engineering and research as VP. Our 10+ person team of software engineers, ML researchers, and human-data experts takes frontier AI-for-science from prototype to production at this post-Google spin-out.
- AI for Science: Worked with perfumers, lab scientists, and annotators to ship cost-and-safety optimization for our formula model, anchoring a multi-million-dollar CPG enterprise deal. Built a GNN end-to-end that predicts intensity from molecular structure, significantly lifting our ingredient-discovery hit rate. Worked with analytical chemists to "teleport" a physical plum scent, translating GC-MS signatures into digital profiles and validating the technology end-to-end.
- Hands-on Agentic AI: Architected Studio, an agentic platform that turns brand briefs into formulas. Frontier LLMs, fine-tuned with SFT and RL, orchestrate our perfumery tools and chemical data through a RAG-like system. An active-learning loop with perfumers feeds an eval stack that lifts model quality release-over-release.
- 0-1 Multimodal ML Stack: Built the post-Google ML stack from scratch (PyTorch, GCP, W&B), connecting olfactory data across molecular structures, fragrance materials, analytical instrument signals, human perception labels, and visual concepts. Shipped training-data pipelines, evaluation frameworks, and deployment infrastructure for the team to build on.
Student Researcher | Google
May 2018 - Sep 2022
- Olfactory ML: Built receptor binding and metabolic activity models that catalyzed Osmo's $60M spin-out. The work led to one utility patent and four publications featured in major news outlets.
- Generative ML for Cell Imaging: First-authored a GAN approach to batch-effect correction in high-content cellular imaging, published in Bioinformatics and open-sourced as part of Google's TF-GAN library.
- Genomics & Structural Variants: Built ML-based filtering for structural variant calling, plus a ~100× faster read-alignment method, filed as a utility patent.
Intern | DeepMind
Sep 2021 - Dec 2021
- AlphaFold Team: Worked with the team behind AlphaFold2 on protein-folding research. Built representation learning models for biological structure prediction in JAX, on top of the AlphaFold2 codebase, at foundation-model scale.
Software Engineering Intern | Uber
Summer 2016 & 2017
- [2017] Sensor & Machine Learning: Built a CRF variant to infer Uber Eats delivery events from mobile sensor data. Won 1st place at Uber's internal ML poster session.
- [2016] Mobile Dev. & Engineering: Built a web-based forensics tool for mobile UI testing that synchronized logs with video timestamps, cutting internal developer debugging time by ~50%.
Education
2017 - 2022
Doctor of Philosophy in Computer Science
Dissertation: Machine learning for drug discovery and beyond (advisor: Jian Peng)
GPA: 4.00 / 4.00 | University Fellowship | Richard T. Cheng Endowed Fellowship
2013 - 2017
Bachelor of Science in Computer Science and Neuroscience
GPA: 3.96 / 4.00 | Summa Cum Laude | Phi Beta Kappa (Junior) | Schiff Fellowship
Services
- Reviewer for Nature Machine Intelligence (2025), ICML (2024), ICLR (2024), NeurIPS (2023), LoG (2022 & 2023), RECOMB (2021), ISMB (2019 & 2020).
- Program Committee for ICML: ML Interpretability for Scientific Discovery Workshop 2020.
Publications (*equal contribution)
A deep learning and digital archaeology approach for mosquito repellent discovery
Chemical Science (2025)
New York Smells: A Large Multimodal Dataset for Olfaction
arXiv (2025)
Pervasive mislocalization of pathogenic coding variants underlying human disorders
Cell (2024)
A Principal Odor Map Unifies Diverse Tasks in Human Olfactory Perception
Science (2023)
3D Equivariant Diffusion for Target-Aware Molecule Generation and Affinity Prediction
ICLR (2023)
Metabolic activity organizes olfactory representations
eLife (2023)
A central chaperone-like role for 14-3-3 proteins in human cells
Molecular Cell (2023)
Energy-Inspired Molecular Conformation Optimization
ICLR (2022)
ECNet is an evolutionary context-integrated deep learning framework for protein engineering
Nature Communication (2021)
Integrating Deep Neural Networks and Symbolic Inference for Organic Reactivity Prediction
ACS National Meeting (2021)
Comprehensive interactome profiling of the human Hsp70 network highlights functional differentiation of J domains
Molecular Cell (2021)
Evaluating Attribution for Graph Neural Networks
NeurIPS (2020)
Batch Equalization with a Generative Adversarial Network
Bioinformatics (2020)
Evolutionary context-integrated deep sequence modeling for protein engineering
RECOMB (2020)