A short list, deliberately. Everything below sits on one loop — monitor a data source, analyze it with AI, alert on what matters. It's the loop our own products run on; anything outside it, we subcontract or refer.
Senior pair-hours for teams that already have one. Source selection, scoring logic, prompt and eval design — the second opinion before you commit to a quarter of work.
A focused dive through the data you want to watch and how you're watching it. You get a written report with prioritised findings — which sources to trust, what you're missing, and where the false positives are coming from.
A pass over how your LLM reads and scores the data — prompts, relevance logic, and the evals (or lack of them) behind it. We tell you where it's hallucinating, where it's miscalibrated, and what to fix first.
For founders & operators sizing the next quarter of work. We'll read your brief, ask the obvious follow-ups, and come back with a one-page plan you can take to your team — or to a different vendor entirely.
Scoped deliverables — a monitoring pipeline, an AI summarization layer, a full data product. Fixed scope, fixed price, weekly demos, written handoff at the end.
The engine behind our own products. We watch a source — a public register, an API, your own database — detect what changed, score it, and deliver alerts your team will actually read. By daily digest or in realtime.
LLMs pointed at your data: plain-English summaries, relevance scores, and structured extraction — grounded in the source, with evals you can actually run. Provider-agnostic across OpenAI, Claude, and Gemini.
The unglamorous plumbing that makes the data — and the alerts on top of it — trustworthy. We ingest from feeds, APIs, and databases, then dedupe, entity-match, and normalize so nothing is double-counted or missed.
An end-to-end monitoring product like the ones we run in-house. Watchlists, scoring, intelligence pages, alerts, and billing — built, shipped, and yours. Something on staging in week one, launch-ready at handoff.
We're explicit about this because it's the fastest way to save everyone a week. If your project is on this list, we know good people; ask and we'll make an intro.
Every project on the same cadence. You'll always know what's next, what we're blocked on, and what's shipping this Friday.
An hour on a call, then a day in your repo and your docs. You get back a one-page plan with a fixed quote and a start date.
A real, working slice — often a first scored record or a sample alert against your own data. A visible stake in the ground that sets the cadence for everything after.
Every Friday: a 15-minute recorded walkthrough of what shipped, what's in review, and what's next. No status decks, no theatre.
Runbook, test suite, and an architecture note you can hand to your next hire. Thirty days of follow-up included, whether you need it or not.
If yours isn't here, send it via the form on the homepage. We answer most questions in plain text the same business day.
Most engagements land between four and ten weeks. A focused piece (a new data source, an AI summarization layer, an alerting flow) is two to four weeks. A full data product is six to twelve. We won't quote anything longer than twelve weeks without a checkpoint in the middle — long projects without checkpoints are how teams get stuck.
Thirty days of follow-up support comes with every project — fixes, small adjustments, a follow-up review. Past that, we offer an hourly retainer at the standard rate. Monitoring systems do need tending as sources change, so a light ongoing arrangement is common — but never an open-ended SLA.
Public registers and open data (USPTO, SEC EDGAR, SAM.gov, USAspending, and the like), third-party and partner APIs, bulk feeds, and your own databases. If a source is reachable and its terms allow monitoring, we can usually ingest, dedupe, and entity-match it. If it can't be done cleanly, we'll tell you before you commit.
Provider-agnostic — the latest OpenAI, Claude, and Gemini models, chosen per task and swappable behind a thin adapter. Every summary is grounded in the source record and linked back to it, and we build an eval harness so you can measure quality rather than trust a vibe.
Small changes (under a day of work) we just absorb. Anything bigger gets written up as a one-paragraph change order with a revised quote, which you approve before we start. No surprise invoices.
Only us. Tiny Prism is a one-person LLC — your data, your accounts, your records stay between us. We sign whatever NDA and data-processing terms you need; we also have defaults if that's faster.
Sometimes. If the badge in the top right says "taking briefs", we have room inside the next thirty days. Either way, we'll quote a realistic start date when you send the brief.
We read every brief and respond within one business day with either a yes, a no, or a referral to someone better suited.