Semantic search across 244 declassified UAP documents from war.gov/UFO — FBI files dating to 1947, DoD mission reports from Syria to the East China Sea, the French COMETA report, and more. Ask in plain English. Runs entirely in your browser.
244 PDFs were downloaded from war.gov/UFO (PURSUE releases 1 & 2). All 122 unique PDFs were processed with EasyOCR on a G4 GPU — pages were first denoised and binarized via Otsu thresholding to separate ink from paper noise before OCR. 86 of 122 documents yielded usable text. The remaining 25 are FBI photograph files with no text content; 2 are heavily redacted forms with minimal recoverable text.
Documents were chunked into ~400-word overlapping segments (2,790 chunks total) and embedded with all-MiniLM-L6-v2 via sentence-transformers on GPU. Embeddings were quantized from float32 to int8 (7.1MB index, cosine error <0.003). Search runs the same model in your browser via @xenova/transformers (WebAssembly) and scores chunks by cosine similarity.
Limitation: FBI-PHOTO-B* files and one composite sketch are photographs with no extractable text and are not searchable here.