Jun 12
これって要するに、「モデルに都度判断させる」から「スキーマの解釈を事前に固定する」への設計シフトだと理解した。
NL2SQLで精度が安定しない原因の多くは、モデルの能力よりも「スキーマの意味が文脈によって揺れる」ことにある。そこを人手で注釈するのではなく、プロービングで自律的に確定させるのが本質的な部分だと思う。地味な仕組みだけど、ここを雑にすると後から確実に効いてくる。
1
6
Jun 12
SOMA-SQLが対象にしているのは、NL2SQLで精度が落ちる3つの曖昧性：
①自然言語の質問そのものの多義性
②スキーマ定義の解釈ブレ
③モデルが推論時に選ぶ解釈のズレ
合成クエリログとプロービングクエリを組み合わせて、スキーマの解釈を事前に確定させる設計。6つの公開ベンチマークで実行精度を平均13.0%改善、曖昧な質問に絞ると最大16.7%向上という結果が出ている。（arXiv投稿段階のプレプリントなので、査読後に数字が動く可能性はある）
1
2
Jun 12
保存しておきたい論文。
NL2SQLの「曖昧さ」問題、ずっと根本解決できないまま運用でごまかしてきた。
SOMA-SQLという手法が、人間の介在なしに自律で曖昧性を解消するアプローチを取っていて、腹落ちする部分が多かった。要点と、次に試すならこう、というのをまとめておく。
🧵 #AI実装ノウハウ
1
3
Jun 8
Looking to scale this work to a full BIRD benchmark evaluation (12,751 queries).
Current result: 60.82% on the BIRD Dev Set using DeepSeek V4 and a custom multi-agent NL2SQL pipeline.
If any model providers or inference platforms are interested in supporting independent benchmark research with API credits or infrastructure, I'd love to connect and share more details.🫶
@deepseek_ai @OpenRouterAI @togethercompute @FireworksAIHQ @GroqInc @xai
#BIRDSQL #TextToSQL #LLM #AIResearch
Jun 8
Built a multi-agent NL2SQL pipeline using DeepSeek and currently achieving 60.82% on the BIRD Dev Set.
Preparing for a full 12,751-query evaluation run and currently limited by API concurrency.
Would appreciate guidance on benchmark-scale quota requests and would be grateful for sponsorship/support for evaluation costs.
@deepseek_ai
1
1
72
Jun 8
Built a multi-agent NL2SQL pipeline using DeepSeek and currently achieving 60.82% on the BIRD Dev Set.
Preparing for a full 12,751-query evaluation run and currently limited by API concurrency.
Would appreciate guidance on benchmark-scale quota requests and would be grateful for sponsorship/support for evaluation costs.
@deepseek_ai
1
98
Jun 7
Project page: sslab-cse-iitb.github.io/tec…
Paper: doi.org/10.1145/3769822
#SIGMOD2026 #Research #TextToSQL #NL2SQL #LLMs #Databases #IITBombay #AppliedAI
10
May 30
这篇论文提出VineLM用Trie结构让代理工作流在运行时动态选择LLM模型，按每次请求的预算（成本/延迟）最大化准确性。
核心：离线阶段将可行路径编码为带注释Trie并稀疏估计准确性、成本；运行时每完成一步后扎根当前前缀，在剩余子树上重新规划。
结果是NL2SQL和数学推理任务准确率提升18%，离线分析成本直降99%，而且边界清晰（只适用可分解的“阶段调用”工作流）。
arxiv.org/abs/2605.23914
2
48
AIで学生指導記録を自動化＆検索！🎓✨
OCIとMySQL HeatWaveの活用事例👀
✅NL2SQLは「ルール＋Few-shot」で精度向上
✅実運用でのリアルな課題も網羅
✅プロンプト設計の重要性を再認識
現場の試行錯誤が詰まった良資料です🧐
speakerdeck.com/heatwavejp/h…
#MySQL #生成AI #AI活用
1
5
723
Considering NL2SQL? Should your database really be the prompt? How can SQL MCP Server help? - from Jerry Nixon
#sqlserver #mcp
devblogs.microsoft.com/azure…
1
1
5
854
May 18
We're excited to reveal CYANSQL, a new Test-Time Scaling framework co-developed with Fudan University! 🚀 Accepted by #ICDE2026, it's a game-changer for complex NL2SQL tasks.
Performance on BIRD dataset:
🏆 87.22% Recall (Industry Leading)
⚡ 73.47% Accuracy
🤖 Powers our TCDataAgent
Check out how we're leading the AI Native big data era! 🌐 icde2026.github.io/accepted-…
#TencentCloud #AI #BigData #NL2SQL #DataScience #LLMs #CloudTech #TechNews #Innovation
1
2
342
Tencent Cloud and Fudan University Propose CYANSQL: A Clustering-based Test-Time Scaling Method to Enhance Complex NL2SQL Generation tencentcloud.com/dynamic/blo…
1
2
2
49
May 9
Generative AI on Google Cloud with LangChain — Design scalable Generative AI solutions with Python, LangChain, and Vertex AI on Google Cloud: amzn.to/4frbkPA v/ @PacktDataML
𝓚𝓮𝔂 𝓕𝓮𝓪𝓽𝓾𝓻𝓮𝓼:
🔴Turn challenges into opportunities by learning advanced techniques for text generation, summarization, and question answering using LangChain and Google Cloud tools
🔵Solve real-world business problems with hands-on examples of GenAI applications on Google Cloud
🟡Learn repeatable design patterns for Gen AI on Google Cloud with a focus on architecture and AI ethics
🔴Build and implement GenAI agents and workflows, such as RAG and NL2SQL, using LangChain and Vertex AI
🔵Purchase of the print or Kindle book includes a free PDF eBook
2
3
9
744
記事を投稿しました！ Oracle NL2SQL Data Retrieval Agentの自律補完を試してみた on #Qiita qiita.com/asahide/items/02e4…
2
140
Did you know you can query a database without writing a single line of SQL?
Just ask:
👉 “Show me last month’s top-selling products”
And NL2SQL turns it into a working query instantly.
We’re building this in-house.
Not just a wrapper around an LLM, but a system that:
• Understands messy business data
• Maps intent → schema
• Handles ambiguity like a human analyst
This is how non-technical teams will explore data soon.
We’re documenting the journey sharing learnings.
If you want early access & behind-the-scenes builds:
Join our Discord ↓
discord.gg/gUz2QzEPK
1
3
54
Apr 29
Generative AI on Google Cloud with LangChain — Design scalable Generative AI solutions with Python, LangChain, and Vertex AI on Google Cloud: amzn.to/4frbkPA v/ @PacktDataML
𝓚𝓮𝔂 𝓕𝓮𝓪𝓽𝓾𝓻𝓮𝓼:
🔴Turn challenges into opportunities by learning advanced techniques for text generation, summarization, and question answering using LangChain and Google Cloud tools
🔵Solve real-world business problems with hands-on examples of GenAI applications on Google Cloud
🟡Learn repeatable design patterns for Gen AI on Google Cloud with a focus on architecture and AI ethics
🔴Build and implement GenAI agents and workflows, such as RAG and NL2SQL, using LangChain and Vertex AI
🔵Purchase of the print or Kindle book includes a free PDF eBook
2
14
70
3,919