Hello Vardhan, Hope you are doing well. Thank you for your PR, we will review it and get back to you.
This is a good direction to explore, and it does align with the project goals at a high level. A small rule-based prototype for simple cases makes sense as a starting point, especially if it helps clarify the kinds of user inputs, query patterns, and failure modes we should expect before leaning too much on an LLM-based fallback for more complex queries. One important aspect to keep in mind is that AsterixDB already supports a large set of built-in functions, including many specialized ones. For an NL2SQL++ system to be genuinely useful, it should be aware not only of the schema, but also of the available functions, what each function expects as input, what it does semantically, and what kind of output it produces. That functional awareness is important so the system can compose correct SQL++ queries and make meaningful use of the engine’s capabilities rather than only generating simple select-filter-project patterns. In terms of next focus, I would suggest prioritizing the *NL2SQL design/scoping* first: - define the supported query classes, - identify what metadata/context the system needs (schema, function signatures, examples, constraints), - and think through validation/correction flow for generated queries. API/documentation improvements are still valuable, but they will likely be more effective once the intended NL2SQL workflow and requirements are clearer. I would encourage you to make the prototype concrete, keep the scope controlled, and explicitly think about how function knowledge will be represented and exposed to the assistant. Best regards, Suryaa On Wed, Mar 25, 2026 at 6:08 AM Vardhan Kumar <[email protected]> wrote: > Hi Surya, > > I’ve been following the recent discussions on the NL2SQL++ project and > started contributing accordingly. > > I recently submitted a PR adding and improving documentation for several > Admin API endpoints: > https://github.com/apache/asterixdb/pull/43 > > Alongside this, I’ve begun exploring the NL2SQL++ direction by implementing > a small prototype using a rule-based approach for simple queries, with the > intention of extending it using an LLM-based fallback for more complex > cases. > > I would really appreciate your feedback on: > > - Whether this direction aligns with the project goals > - What would be the best next area to focus on (API documentation > improvements or NL2SQL design) > > Thank you for your guidance. > > Best regards, > Vardhan Kumar >
