Common Bill

Inspiration

An independent cinema and a bookshop may share a neighborhood but still struggle to agree on an evening they can run together. A clever thematic pairing is only the beginning: someone must check the budget, screening time, stock, availability, and who will host the discussion.

Common Bill turns that coordination problem into a shared planning desk: two businesses, one evening worth doing together.

What it does

The agent reads both briefs, filters infeasible film–book combinations, selects an evaluation scope, queries Qloo in both directions, and prepares an evidence-backed program with a running order, host brief, and announcement draft. Both partners can record local reviews. Reject a pairing or lose the bookseller, and the agent revises the plan while preserving hard requirements. A before/after view shows what changed.

Try the public live demo without signing in or supplying a key. Its cultural evidence comes from real Qloo calls and its planning uses Groq. Business names, costs, stock, permissions, and staffing are explicitly fictional demonstration assumptions—not verified partner facts. Results may be reused for one hour with their original timestamps.

Why Qloo matters

An LLM can write a persuasive connection between almost any film and book. Common Bill asks for evidence alongside that interpretation. It resolves exact works, constrains Qloo Insights to the available catalog, and queries film-to-book and book-to-film affinity separately. A Qloo-mode proposal needs both directions. Missing evidence remains unknown.

Every decision retains the entity IDs, query parameters, vendor scores, candidate ranks, timestamps, feasibility checks, and a short editorial explanation. The app's relative-rank heuristic is explicitly distinguished from Qloo affinity; neither is presented as audience overlap or demand.

The evidence test compares Qloo + LLM with the same LLM using identical operational inputs. Display order is randomized, methods are initially hidden, and no preference is a valid result. Removing Qloo removes the evidence gate and provenance; it does not magically make the baseline incapable of forming an editorial opinion.

How we built it

TypeScript, React, Next.js, Tailwind CSS, Radix UI, and Framer Motion power the interface. Server-side Qloo adapters handle search and Insights. Groq GPT-OSS 120B returns structured planning decisions. A deterministic constraint engine rejects ineligible choices. The public deployment runs through Sites using Vinext and Cloudflare Workers. Complete source, setup instructions, tests, and an MIT license are available on GitHub.

Challenges and what we learned

Ambiguous editions require exact identity checks. Affinity is directional, and a singleton result provides no comparative ranking advantage. Provider rate limits require compact context, bounded retries, and honest error states. Most importantly, a culturally attractive plan must still be operationally feasible.

What we verified

All 21 regression tests and the production build pass. A dated live integration check exercises initial planning, staffing repair, rejection, infeasible budget, and the model-only baseline. These checks establish implementation behavior using illustrative inputs—not real-world impact.

Potential impact and what's next

Our hypothesis is that a shared evidence-backed draft can reduce coordination effort for cinema programmers and booksellers. We have not measured time savings, attendance, revenue, or adoption, and claim no organizer endorsements. The next step is a consenting operator pilot using verified catalogs, blind proposal preference, task completion, and planning time. A pilot protocol is included in the repository.

This release is a single-device planning pilot. Reviews are local notes, not authenticated approvals; nothing is automatically promoted, booked, or purchased.

Try it in three minutes

1. Click Try the live demo.
2. Open Inspect the evidence.
3. Record both partner reviews, then try Bookseller unavailable.
4. Try Budget drops to £100 to see an honest stop.
5. Reset the scenario and run The evidence test.
6. Export the event brief and decision receipt.

Built With

• qloo
• typescript