Research vault for vetted, citable data points
Build a "research vault" — a living, curated library of citable data points (stats, survey findings, benchmarks, quotes) for a marketing/content team. The problem it solves: people half-remember a statistic they saw somewhere and can't find the source. This vault holds only data points a human has vetted as relevant and reputable, and lets anyone retrieve them with sources in seconds — the key difference from just asking a language model. Provide a single web page (an HTTP route) as the main entry point, with two modes: one to contribute a data point and one to ask the vault a question. Keep it private to your team by default. Contributing: a person submits a data point with its source and optional context. Use an AI model to clean it into one quotable sentence, infer a readable source name, assign topic tags for filtering, and identify how recent the data is (the year/period it refers to). Store each entry in durable storage that persists and grows over time. Also allow contributions from your team chat — e.g. a Slack slash command, or whatever chat tool your team uses — so people can file findings without leaving the conversation. Querying: a person asks a natural-language question. Load the stored data points and use an AI model to return the ones that genuinely answer it, each shown with its source, recency, and tags. If nothing in the vault matches, either say so clearly or answer from the model's general knowledge — but flag that answer unmistakably as NOT coming from the vetted vault so it's never mistaken for a citable source. Always show a clear response, including when there are no matches. Triggering is entirely on-demand via the web routes and the chat command; there is no schedule or upstream dependency. Use an AI provider of your choice (I used Anthropic's Claude — swap in OpenAI, Gemini, or similar), post/receive from Slack or your team's chat tool, and keep the data in the platform's built-in persistent storage rather than an external database. Optionally provide a bulk-import path to seed the vault with an initial batch of data points. Tools used: 3B React page (web UI), 3B HTTP routes, 3B named volume storage, Anthropic Claude API (via connector), Slack (slash command)
What this prompt builds
The Citable Research Vault lets content and marketing teams submit and retrieve vetted data points—statistics, survey findings, quotes—through a web page or Slack. It uses AI to clean, tag, and make each entry searchable by natural language, ensuring only credible, recent sources are returned. The workflow eliminates duplicated research effort and raises the quality of team storytelling by centralizing vetted data in persistent storage.
The problem
Content and marketing teams struggle to find relevant, recent research from reputable sources—asking LLMs ad hoc returns mixed, unverified results, and there's no way to share data points the team has already vetted. Useful statistics surface informally in Slack or email and then vanish, forcing colleagues to duplicate research effort and miss opportunities to use the strongest possible data in their storytelling. The Citable Research Vault solves this by letting the team submit vetted data points—from reports, surveys, or news stories—via a web page or directly from Slack into a centralized, persistent vault. Anyone can then query the vault with a natural-language question and instantly retrieve only data that's been checked for relevance, recency, and source credibility, complete with citations.
Solution and impact
The workflow saves the entire content and comms team time by eliminating duplicate research, increases the likelihood of finding the best data for any story, and improves the overall quality and credibility of the organization's storytelling. Each contribution is cleaned by AI into a quotable sentence, tagged by topic, and timestamped for recency, ensuring queries surface only vetted, citable sources. When no match exists in the vault, the workflow clearly flags any general-knowledge answer as unverified, preventing teams from mistaking non-curated information for a credible source.