What you’re looking at

The Lexington Times is three products pretending to be one paper. Here’s what each one does and where it runs.

The stack

Public data sources

Everything below is either public record, a licensed re-publisher, or data we generate ourselves. No scraping of paid sources. No social-media private APIs. No PII.

The meeting pipeline

LFUCG publishes Council + commission meetings on Granicus with scanned agenda PDFs and auto-generated captions. That’s where our civic coverage starts — but the raw feed isn’t something you can skim. A dedicated Python pipeline turns each clip into something searchable.

The whole thing is open source — code, docs, and the Granicus reverse-engineering notes (granicus.md) that made timestamp deep-linking possible.

What LexBot reads on air

Every voice segment is a written-for-radio script generated by Claude, spoken by ElevenLabs, and logged to the transcript archive. The cadence is deliberately unhurried — it’s a civic radio feed, not TikTok.

Guardrails

• AI disclosure — every transcript includes a human-readable note that the segment was generated by LexBot, which model wrote it, and when it aired.
• Fair use — aggregated headlines carry title + ≤200-char summary, with the canonical source link. Voice scripts paraphrase and attribute; they never quote.
• No hallucinations on facts — Ask Lex and breaking briefs run through Anthropic’s web_search_20250305 tool, geo-biased to Lexington, KY.
• Circuit breakers — every poller opens on 3 consecutive failures and stays open 5 minutes before retry. No silent infinite-retry.
• Dwell + cooldown — scene-switch safety constants prevent the stream from rapidly flipping between cams / alerts / ads in ways that would look unstable on air.
• Fourth wall — hosts never say “I don’t have that in front of me,” “our data,” or “checking our feeds.” If information is thin, we pivot to what a local would know.
• Crime blotter is aggregated — counts per council district and crime type. We don’t name individuals or addresses, even when the public data set would allow it.
• YouTube quota safety — live-title and chat polling are rate-tuned to stay under the 10k-unit/day Data API cap. Never auto-publishes without the stream-key gate.
• Human in the loop for Shorts — highlight clips go through a two-step unlisted-upload → manual review → publish flow.

Humans in the loop

There is a human behind all of this. Most of it was built, is operated, and is continuously debugged by Paul Oliva, with help from friends and open-source contributors. There is an ops control panel on :3008 with one-click overrides for:

• skipping / pinning a segment on the podcast rotation
• forcing a specific camera in the 2×2 tile mode
• injecting weather alerts, breaking headlines, and incidents for testing
• rotating the Facebook broadcast token, cutting over to a new live_video
• approving viewer-submitted milestones + shoutouts
• pinning a civic graphic on-air for a fixed dwell

The paper will only ever be as opinionated as the humans holding the ops panel.

Questions we get

This is pretty cool. Can we hire Paul?

Paul M. Oliva — Senior Full Stack Engineer

Yes — he is open to new opportunities. Paul built the paper, LexBot, the meeting pipeline, and most of the civic-data plumbing on this site on top of a full-time engineering job — shipping features end-to-end across a Next.js frontend and a Node/TypeScript API for a high-volume ticketing platform.

What you’re hiring, in short:

• FrontendReact, Next.js, TypeScript, Redux, Tailwind
• BackendNode.js, Express, REST, GraphQL, tRPC, FastAPI
• AI / LLMOpenAI, Claude, RAG, ChromaDB, Tesseract OCR
• CloudAWS (Lambda, S3, RDS, SNS/SQS), Docker, GitHub Actions
• DataPostgreSQL, MySQL, DynamoDB, MongoDB, Redis
• PaymentsAdyen, Stripe, PayPal
• ObservabilityDatadog, Sentry, CloudWatch

pauloliva.com →