Get Full Government Meeting Transcripts, Videos, & Alerts Forever!
Get email alerts on the Ai For Msrs topic
No spam. Unsubscribe anytime.
LAFCO hears AI vendor pitch to speed municipal service reviews; commissioners ask for pilot study
Summary
A vendor presented a Napa LAFCO pilot using retrieval‑augmented AI to assemble municipal service reviews quickly and cheaply; commissioners expressed interest but asked staff to explore a limited pilot rather than commit to a multi‑year contract.
David Key, founder and CEO of the Paperworks Company, demonstrated an AI‑driven workflow the firm used in a Napa LAFCO pilot to process agency documents and generate a municipal service review draft in under five weeks. Key said the Napa pilot produced a detailed draft for roughly $5,000, compared with a conventional consultant contract that can cost six figures and take many months.
The technology Key described uses a retrieval‑augmented generation (RAG) architecture that searches a sandboxed, jurisdiction‑specific document store rather than querying public models, and Key emphasized editorial controls: human review and line‑by‑line citations to source documents to reduce the risk of AI hallucination.
Commissioner response and concerns: commissioners thanked the presenter but raised practical questions: the quality and completeness of underlying agency data ("garbage in, garbage out"), how to handle data gaps, expected costs if Santa Barbara LAFCO asked the firm to update an entire catalog of MSRs (a quoted illustrative flat fee of about $175,000/year in a multi‑county contract), and whether the technology would displace staff. Several commissioners and executive staff favored a phased approach — a narrowly scoped pilot or one‑off use case for a single MSR or staff report — rather than an immediate two‑year commitment.
Staff direction: Executive Officer Mike Prader and multiple commissioners asked staff to follow up with the vendor on a succinct pilot proposal and cost estimate, to identify which local MSR(s) would be the best first test, and to return to the commission before any contractual commitment.
What was shown: Key walked through three risk categories (hallucinations, contextual errors, and data‑quality gaps) and how RAG plus human editorial review mitigates them. He also showed audit‑style inline citations in the Napa draft and suggested the tool could be reused as an institutional memory vector database to make prior reports searchable.
Outcome: the commission received the demonstration and directed staff to report back with a more detailed, scoped recommendation (pilot parameters, deliverables, budgetary impacts) before any engagement. No contract was approved at the March 5 meeting.
