OpenAI Researcher Audrey Cheng Joins Stanford, Advancing AI-Driven Systems Research
Chinese-American researcher Audrey Cheng, a visiting researcher at OpenAI, will join Stanford's Computer Science department as an assistant professor in fall 2027, bringing a research portfolio spanning database optimization for Meta's TAO and multi-tenant scheduling to pioneering AI-driven systems research that automates algorithm discovery for tasks like cloud scheduling and LLM inference.
Background and Academic Path
Audrey Cheng completed her undergraduate studies at Princeton University and began her Ph.D. in Computer Science at UC Berkeley in 2020, advised by distributed systems experts Ion Stoica and Natacha Crooks. Her research centers on database system performance optimization , addressing how to handle billions of concurrent requests while maintaining data correctness and stability across read/write, transaction processing, and resource scheduling.
Key Research Contributions
RAMP-TAO: Consistency for Meta's TAO Store (VLDB 2021)
As first author, Cheng published RAMP-TAO at VLDB 2021, winning the Best Industry Paper Award. The work targets Meta's online data storage system TAO, solving the challenge of preventing users from reading incomplete data when multiple updates occur simultaneously. The paper demonstrates how to achieve atomic visibility for multi-object transactions without sacrificing latency.
Take Out the TraChe: Cache Optimization (OSDI 2023)
In 2023, Cheng co-authored Take Out the TraChe , accepted at OSDI and recognized with the Meta Next-Generation Data Infrastructure Research Award. This research improves database runtime efficiency by optimizing cache management to boost read performance for frequently accessed data.
DRFT: Fair Multi-Tenant Scheduling (VLDB 2025, Best of VLDB 2025)
In 2025, Cheng tackled multi-tenant database resource allocation with the DRFT scheduling algorithm . The problem: when multiple enterprises share a database, some tenants may monopolize resources, degrading performance for others. DRFT improves fairness across tenants while limiting system throughput overhead to under 5%. This work was selected as a Best of VLDB 2025 paper.
AI-Driven Systems Research (ADRS)
Moving beyond hand-crafted optimizations, Cheng has recently pioneered AI-Driven Research for Systems (ADRS) , introduced in the 2025 paper Barbarians at the Gate . The ADRS framework automates the algorithm research loop: an AI agent generates multiple algorithm candidates, automatically runs empirical tests, selects the best performers, and iterates further. This shifts AI from a coding assistant to an autonomous algorithm explorer.
Applications and Results
The team applied ADRS to four domains:
Multi-region cloud scheduling
Mixture-of-Experts (MoE) inference
LLM-based SQL query optimization
Transaction scheduling
In several cases, AI-discovered algorithms achieved up to 5× runtime improvement or 50% cost reduction compared to human-designed baselines.
Follow-up Projects: AdaEvolve and EvoX
Cheng further explores combining large language models with evolutionary search in AdaEvolve and EvoX , enabling AI to not only propose optimizations but also continuously refine its search strategy based on experimental feedback.
Recognition and Industry Impact
2024 EECS Rising Stars
NSF Graduate Research Fellowship (GRFP)
Meta PhD Research Fellowship
Berkeley Chancellor's Fellowship
Her database research has been adopted in production by Meta, PlanetScale, and TiDB .
Next Chapter: Stanford Faculty
Starting fall 2027, Audrey Cheng will join the Stanford Computer Science Department as an assistant professor, continuing to push the intersection of AI and computer systems research—advancing AI from a coding aid toward autonomous algorithm discovery and system optimization.
Signed-in readers can open the original source through BestHub's protected redirect.
This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactand we will review it promptly.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.
