An AI-powered research database for the UFO and UAP phenomenon
UAP Files is an AI-powered research database built for anyone investigating the UFO and UAP phenomenon. Every video in our archive is automatically transcribed using AI speech recognition, then analyzed with natural language processing to extract named entities — the people, organizations, locations, dates, and key terms mentioned throughout each conversation.
The result is a fully cross-referenced database where you can trace a name like Bob Lazar or an organization like the CIA across every video in the archive and jump directly to the exact moment they're discussed. Each timestamp is clickable, linking you straight to that point in the original YouTube video.
This site is a demonstration of how AI can transform long-form video content into a structured, searchable research tool. Because transcriptions are generated entirely by AI, you may notice occasional errors in spelling, names, or phrasing. Manually correcting every transcript across hundreds of hours of video isn't feasible for a solo project, but the core technology speaks for itself.
For content creators, this kind of system unlocks enormous value. Every word you've ever spoken on camera becomes searchable, cross-referenced, and timestamped — a living catalog of your entire body of work. Need to find the exact moment you discussed a topic three years ago for a callback in your next video? It's one search away. Want to pull clips across dozens of episodes for a highlight reel or marketing cut? Every mention is indexed to the second.
Beyond production, it gives creators and their teams a structured backend — a way to see who and what you've covered, how often, and when — turning hours of unstructured content into organized, actionable data. Whether you're a solo YouTuber, a podcast network, or a research organization, this is what your content library could look like.
When new videos are added to the archive, they go through an automated pipeline that handles everything from start to finish. The system monitors a curated YouTube playlist, and when new videos appear, it downloads the audio and generates a full word-for-word transcription with timestamps. From there, the transcript is scanned to identify every person, organization, location, date, and key term mentioned. Each of these entities is then cross-referenced against every other video in the archive, building the web of connections you see throughout the site. Finally, an AI-generated summary is created so visitors can quickly understand what each video covers before diving in.
The entire process — from new video to fully searchable, cross-referenced entry — runs through a single command. Every transcription, every entity, and every timestamp is generated automatically without manual intervention.
This site runs entirely on a local machine — a personal workstation handling the transcription, entity extraction, and summarization. It works, but it has the limitations you'd expect from a proof of concept. Transcriptions are generated by open-source AI models running locally, which means occasional errors in spelling, names, and phrasing that would need manual review to catch. Entity extraction picks up most references but can misidentify words or miss context. There's no real-time processing, no user accounts, and no editorial workflow for cleaning up results before they go live.
A production version of this platform — built for a content creator, media company, or research organization — would be a fundamentally different product. Transcription would run through OpenAI's Whisper API on high-performance cloud infrastructure, delivering faster and more accurate results. A dedicated review layer would allow editors to correct transcription errors, merge duplicate entities, and curate the data before publishing. The entity extraction would use fine-tuned models trained on the client's specific domain, dramatically improving accuracy for names, terminology, and jargon unique to their content.
Beyond accuracy, a paid platform would include features like automated clip generation from timestamp data, integration with video editing workflows, real-time processing as new content is published, custom dashboards for tracking content trends over time, and API access so the data can feed into other tools and platforms. The infrastructure would scale to handle thousands of hours of content with enterprise-grade reliability.
What you see on this site is the foundation — the core concept working end to end. The jump from proof of concept to production platform is a matter of infrastructure, polish, and customization to fit a client's specific needs.
Interested in building something like this for your own content? Reach out at [email protected]