Alnur Ali Email & Phone Number

San Francisco, Ca, Us

Leading projects applying large language models / generative AI to enterprise data, at Databricks

Greenwich, Ct, Us

I advise / consult for various machine learning startups (typically lying within the span of my social network). I love helping people think through (technical) problems – so if you’d like to chat, please feel free to reach out!

San Francisco, California, Us

(Sisu was an a16z-backed analytics startup, which was acquired by Snowflake)• Managed teams of 10-20 research scientists, engineers, product managers, and designers working on (i) large language models / generative AI; (ii) time series analysis; (iii) segmentation analysis; and (iv) causal inference applied to enterprise data — leading to significant YoY MAU growth. The models we built were based on my work on responsible AI / anomaly detection at Stanford.• My team also helped build out our machine learning infrastructure (using Ray) — leading to a 10x increase in the number of models shipped• Helped define the overall company strategy, engineering roadmap, etc. Helped close several hundred million dollar deals.• Technologies: cloud data warehouses (Redshift, BigQuery, Snowflake, …), cloud storage, Parquet, Arrow, React, Flask, Ray, Kubernetes, Python, Rust

Stanford, Ca, Us

• Worked on "responsible AI" / "AI safety" / reliably and safely deploying ML models• Built algorithms that (i) track deployed model performance, raise an alarm when performance degrades, and generate interpretable diagnostics; (ii) repair and protect any deployed models from future degradations in performance without retraining; (iii) train models without carefully tuning step sizes (see http://www.alnurali.com/ for more info)• Mentored 5 senior Ph.D. students• (Some) technical keywords: distribution shift, robust optimization, subpopulation-level performance, conformal inference / distribution-free uncertainty quantification, weak supervision, tuning parameter-free stochastic optimization, heterogeneous treatment effect estimation, time series analysis, representation learning / embedding• Application areas: finance, operations research, public policy, epidemiology

• Developed and productionized high-risk research ideas on ML for web search• Mentored 5 research interns• Technical keywords: personalized click prediction, localized ranking for better subpopulation-level performance, "early exits" for faster gradient-boosted decision trees

Redmond, Washington, Us

• Worked on ranking and query classification / reformulation / rewriting• Led teams of 5-10 engineers and research scientists to (i) build and ship models; (ii) do feature engineering; (iii) mine terabytes of user behavior logs; (iv) run A/B tests — leading to (roughly) 20% improvements in Bing's accuracy. For these improvements, I was given a Gold Star Award (Microsoft’s discretionary award for technical contributions) 3 times, and granted 8 U.S. patents.• Significant experience (i) defining product strategy and engineering roadmap; (ii) working cross-functionally to get things done; (iii) building high-performing teams; (iv) evaluating acquisitions• Tech transfer to bring research ML into the enterprise (Sharepoint, Office, Dynamics 365, Power BI)• Responsible for a number of 1sts: (i) led the 1st integration of statistical machine translation into a commercial ranking system; (ii) built the Bing team’s 1st active learning pipeline; (iii) built Sharepoint’s 1st data-driven auto-complete system• (Some) technical keywords: gradient-boosted decision trees, neural nets, contextual bandits, off-policy evaluation, robust ranking models, random walk-type representations, natural language processing, sequence models (hidden Markov models, conditional random fields, …), MapReduce

Postdoctoral Scholar, Departments Of Statistics And Electrical Engineering

Doctor Of Philosophy - Phd, Machine Learning

Bachelor Of Science - Bs, Computer Science