Jiale Lao (@lojil192574) profile picture

I am a CS PhD student at Cornell University. I am interested in leveraging advanced techniques from LLMs to enhance database systems.

Ithaca, NY (USA)
Joined October 2023
Photo by Jiale Lao (@lojil192574)
For over 50 years, we build systems by hand. Is that still the right way in the LLM era? With a few thousand dollars, GenDB-V can build a database engine, specialized to your workload and hardware for performance, and formally verified for correctness. genlang-v.github.io/GenDB-V/
6
1
3
17,989
Two lessons from the history: (1) Performance depends on the workload and the hardware. Both change over time, and adapting to changes means redesigning and rebuilding systems at great expense. PostgreSQL, Snowflake, DuckDB, Sirius, you name it. This cycle will keep going.
1
161
Welcome to GenDB! solidlao.github.io/GenDB/# GenDB generates a query engine on the fly for your specific data, workloads, and hardware. 1 hour, $30: 3-7 times faster than DuckDB, 282-462 times faster than PostgreSQL. Using Claude Code or Codex with latest models: Opus 4.6, GPT-5.4, Sonnet 4.6, GPT-5.3-Codex. Tried different programming languages: C++ and Rust. #LLMs #Database #Claude #GPT #CPP #Rust #DuckDB #PostgreSQL
💡 Two arXiv papers published in recent days (one from us, one from TUD) reach the same conclusion: LLMs can now generate C++ code for SQL processing that outperforms classical database systems. ⚙️ Our code generator is based on Claude Code and exploits multiple agents working in parallel. Each agent performs tasks typically associated with different components in a #DBMS, such as workload analysis, query optimization, or physical design tuning. 📊 We compare to various classical #DBMS such as DuckDB, ClickHouse, Umbra, MonetDB, and PostgreSQL, finding that the agent-generated code is often significantly faster. Code generation costs are moderate (<$20), making the approach practical for frequently executed queries. 🤖 Analyzing generated code, we find that agents exploit various optimization techniques, including query-specific data structures, as well as low-level optimizations that are specific to the hardware cache hierarchy of our server. 📃 Paper: arxiv.org/pdf/2603.02081 💾 Code: github.com/SolidLao/GenDB 🌐 Site: solidlao.github.io/GenDB @lojil192574 #LLM #Databases #AI #DB
420
We just publish a project, GenDB, that uses multi-agents (Claude Code or Codex) to synthesize instance-optimized execution code for each SQL. With feedback-driven iterative optimization, GenDB is 2.8x faster than DuckDB/Umbra on TPC-H. More details: github.com/SolidLao/GenDB.
1
326
Excited to share that (1) GPTuner wins SIGMOD Research Highlight Award 2024🏆, (2) SQLBarber is accepted for demonstration at SIGMOD 2025👏! sigmodrecord.org/sigmod-reco… github.com/SolidLao/GPTuner youtube.com/watch?v=qOuAKVXX… @SIGMODConf #LLMs #DB #AI #Data
SIGMOD Record - Issue Mar 2025 (Vol.54,No.1) : sigmodrecord.org/sigmod-reco… It's a special issue presenting papers & tech. perspectives covering a selection of the best conf. papers from 2024 in data management. These papers received the '2024 ACM SIGMOD Research Highlight Award'.
1
246