IPTD-480
Ask NJEMS- natural language queries of NJEMS
Description
This project would pilot the use of Large Language Models (LLMs) to use natural language prompting to produce SQL queries that return data and execute data analysis. This will require pairing one or more LLMs tuned or trained in databases and SQL, with NJEMS database schemas and metadata in order to enable model outputs of figures, tables and possibly charts.
Problem Statement
Too much DEP data remains under-used. One obstacle to greater use has been the friction involved in learning to use a reporting tool or engaging with report experts.
Project Justification
Unlocking greater insights from existing data is likely to directly advance NJDEP goals. Constraints in personnel time and limited report building expertise remain a likely obstacle to rapid and iterative querying that leads to insight.
Estimated Transactions
None
Target Rollout Date
None
Target Rollout Date Reason
None
Activity
Show:
Added TIP docs (see attached) from March 30 - rec’d 2026-04-02
Request # 06440781
Project Name: Ask NJEMS - natural language data retrieval
Agency: ENVIRONMENTAL PROTECTION
Organization: ADMINISTRATIVE OPERATIONS
Agency Point Of Contact: Knute Jensen
The Technology Initiation Proposal (TIP) review for the above project was held on 03/30/2026 and thank you to all who participated.
Attached are the discussion points and action items from the review (see RECAP Excel spreadsheet) as well as a copy of the TIP MARKUP (see TIP PDF document), which includes any updates made during the TIP meeting. Please notify sar@tech.nj.gov if there is any incorrect or missing information.
NOTE: The Agency project team should update progress on any action items in the attached RECAP Excel spreadsheet and submit it back to SAR@tech.nj.gov for the next SAR phase.
WHAT’S NEXT: The next SAR phase is LSAR/C-LSAR, so please download the applicable document from https://tech.nj.gov/it/whatwedo/sar/ and submit the completed document along with your updated RECAP spreadsheet to SAR@tech.nj.gov when ready.
The LSAR/C-LSAR stage is not yet complete in SimpliGov.
TIP Meeting Files:
RECAP Spreadsheet: DEP Ask NJEMS Natural Language Data Retrieval 20260330 TIP RECAP.xlsx
TIP MARKUP: ENVIRONMENTAL PROTECTION Ask NJEMS - natural TIP Request 06440781
Additional documents:
Latest in-house efforts include thorough review of the CGI work and approach which is now outdated and may have been off the mark. Shyam is working toward the context we will need for understanding the tables and schema. He has built a JSON with a schema interpreted through AI that had access to the entire database metadata. We are seeking info from Jim Bridgewater that is manually compiled about the various tables, their naming, prefixes, and uses. We expect to leverage both sets of info in a pre-run loop to find/confirm the subject matter domain(s) implied from any question that will set which tables end up needed. Manav and Shyam are working on multiple ways to architect the whole tool and will be able to run all of them to see what works, since we can use the in-house AI machine without burning token costs. I am seeking volunteers for their approach to pick a limited set of tables for an initial run where we can efficiently compare the architectures in one domain (like hazardous waste or enforcement). We will need program volunteers who can confirm readily if queries are accurate or not. No point working with domains without ready volunteers.
Shyam/Mike McCormack got access to CGI AI server to do work. Mike is putting together a list of what we need from CGI to transition remaining work to in-house development
CGI Pilot Completed. Further work on AskNJEMS will be done in-house.
Work Plan 187 signed 1/24/2025