Municipal Intel

An independent research and engineering project that reads municipal documents, public meeting records and local government budgets at scale and shows what a municipality is planning, funding and leaving unfunded, with the source and the date on every line, across 1.9M+ public funding records.

Independent work, built on my own time from public records.

What a municipality is planning is public, but it is scattered across thousands of agency websites, scanned PDFs, meeting recordings and dozens of funding programs, and most people only find out once a decision has been announced.

Research software that reads what a municipality publishes, budgets, council minutes, meeting recordings, grant awards and announcements, and shows its strategic goals, what it is funding and what it is not. It joins what an agency needs, the money it has and the people who decide into one cited record, for anyone who wants to understand what their municipality is doing.

What a record should tell you

One record puts an agency's stated need, the money behind it and the people who decide into a single view, with the source and date on every line, so whoever is reading can check the evidence before acting on it. A funding award, a budget line and a grant the agency might qualify for are three different things; the record keeps them apart, and it says whether a connection is documented or inferred.

Design rule: Every claim points to a public source and a date. What you see is the raw evidence, dated: name, role, department, source. There are no confidence scores and no fit scores, because a score hides the reasoning you need to check.

Example Township

Municipality · Example County

Need:: Council minutes and the meeting recording both discuss replacing connectivity for field crews.
Funding:: Award listed in a state grant program record.
Decision-maker:: Jordan Example · IT Director · Department of Public Works
Source:: Example Township council minutes, the meeting recording, transcribed, and the staff directory page
Date:: 2026-01-15

Illustrative record. The names are placeholders. This shows the shape of a record; a real, cited one takes its place when I export it. Cited, no score.

Scale

1.9M+ funding records, from 100+ sources.
100+ sources, in different formats.
2,000+ municipalities, seeded across seven states: the six New England states plus Pennsylvania.
500+ concurrent workers, local and cloud capacity; a whole state seeds in a single run.

These figures count funding records, sources and seeded municipalities, not users or customers, and seeded is not the same as fully processed.

Where Municipal Intel holds public records

Each square is a state. The hatched ones are where Municipal Intel holds records today.

Municipal Intel data

Hover, focus or tap a hatched state.

How it works

Why it's hard: The sources come in 100+ different formats, the records don't share identifiers, a lot of it is scanned paper or audio that has to be read or transcribed before it means anything, and a whole state has to be rebuildable on demand. I built it end to end, with AI coding agents writing most of the code under my direction.

Sources
100+ public sources in every format: web pages, scanned PDFs, meeting minutes and meeting recordings.
Ingest
No LLM
No model is involved here, on purpose: this part has to be deterministic, auditable and cheap to run again. 500+ workers launch every county at once.
Extract
Local GPU
Scanned pages get read with OCR. Recordings get transcribed, each speaker identified, on a GPU in my office, so there are no per-minute fees.
Unify
Stitches records that share no identifier into one picture: what the agency needs, the money, and the people who decide.
Retrieve
Full precision
Search by meaning and by keyword at once, on full-precision embeddings in PostgreSQL with pgvector, and every hit carries where it came from.
What you see
What you see is the raw, dated evidence: name, role, department, source and date, with no scores attached.

Six steps, top to bottom. The drawing follows the list as you scroll.

Trace a field back to its source

Pick a field to see where it comes from.

Illustrative. The drawing shows how one record moves, not how many there are. The 100 source slots and 500 worker cells are drawn at the floor of the counts above.

Stack and how it's built

Python and FastAPI on the back end, PostgreSQL with pgvector (managed in production), Next.js in front. Search works by meaning and by keyword at the same time, and every result carries where it came from. Each workspace can see only its own rows, and operators get a read-only MCP server.

AI coding agents built most of it, under written architecture decisions and verification gates. The simplest of those gates: every database write gets read back before the job counts as done.

Where I use AI, and where I don't

Five decisions in the running system, each made on cost, on whether I can audit it afterward, and on quality.

Where I use AI, and where I don't
Job	Choice	Why	Effect
Ingesting 1.9M+ records	No LLM	WhyMust be deterministic, auditable and cheap to re-run.	EffectThe full dataset re-runs the same way every time.
Transcription and speaker diarization	Local GPU	WhyPer-minute API pricing adds up at volume.	EffectNo per-minute API fees.
Agent reasoning	Hosted + local	WhyA hosted frontier model does the thinking: it plans, reviews, edits and evaluates. Local models do the volume work, inside harnesses, evals and tests that run on my own machine.	EffectFrontier model for planning and review; local models for the volume.
Embeddings	Full precision	WhyI don't trade retrieval quality for storage savings.	EffectUsed for every embedding.
Statewide runs	500+ workers	WhySerial runs are too slow for statewide coverage.	EffectA whole state seeds in one run.

My contribution

The concept, the requirements and the architecture are mine, and so are the checks that decide whether a job's output is good enough. AI coding agents did most of the implementation under my direction. What I brought that an engineering team alone would not have is years of working with government: I know what a reader needs to see in a record, and what a wrong inference costs them.

More of my applied AI work The public-sector record