The $0.45 Face Scanner

# The $0.45 Face Scanner

I'm face blind. Clinically. Prosopagnosia. I can't tell you who's sitting across from me at dinner if I haven't heard their voice yet.

So when the DOJ released 329,474 documents in the Epstein files — including 42,182 scanned evidence photos — the idea that I would personally review them for identifiable faces was absurd.

I built a machine to do it instead.

The Pipeline

**Step 1: Face Detection** — MediaPipe BlazeFace running across 42,182 JP2 images on a MacBook M3. Six CPU cores, 94 minutes. Found 279 images containing faces (297 total faces — 18 images had two).

**Step 2: Face Cropping** — Extracted each face using the bounding box coordinates, 30% padding for context. 297 individual face crops.

**Step 3: Reverse Image Identification** — Google Cloud Vision API web_detection on every crop. 297 API calls. Total cost: approximately 45 cents.

**Step 4: Human Review** — Me, a face-blind person, cross-referencing the machine's identifications against our 329,474-document search index.

Total time: about 2 hours. Total cost: $0.45.

Who We Found

Google Cloud Vision identified the following individuals in photographs from the DOJ Epstein evidence files:

| Person | EFTA Evidence | Context |

|--------|--------------|---------|

| **Jeffrey Epstein** | 60+ pages | Throughout — his documents |

| **Ghislaine Maxwell** | EFTA00003712, EFTA00003866+ | Multiple photos, including with identified associates |

| **Jean-Luc Brunel** | EFTA00003878, EFTA00003879, EFTA00003883 | Model agent. Died in custody, Feb 2022 |

| **Les Wexner** | EFTA00003437, EFTA00003591 | L Brands/Victoria's Secret founder |

| **Prince Andrew** | EFTA00003635, EFTA00003791 | Duke of York |

| **Richard Branson** | EFTA00000515 | With Epstein, tropical location. 390 documents of correspondence in index |

| **Brett Ratner** | EFTA00003983 | Film director |

| **Boris Nikolic** | EFTA00003872 | Gates Foundation science advisor, named alternate executor of Epstein's will |

| **Peter Mandelson** | EFTA00003935 | UK Labour politician, Baron Mandelson |

| **Danny Hillis** | EFTA00003422 | Computer scientist, co-founder of Thinking Machines |

| **Tommy Mottola** | EFTA00003898 | Former Sony Music chairman |

| **Anibal Acevedo Vila** | EFTA00003660, EFTA00003661 | Former Governor of Puerto Rico |

| **Marci Hamilton** | EFTA00037120 | Screenshot of CHILD USA founder's bio — Epstein was monitoring the leading expert on child sex abuse statutes of limitations |

| **Ricardo Mangue Obama Nfubea** | Manual ID | Vice President of Equatorial Guinea |

| **Kevin Spacey** | Manual ID | Actor |

Plus 21 high-confidence unidentified faces from what appears to be a personal photo album (pages 6353-6982) that need further identification.

What the Machine Missed

Google Cloud Vision does web matching, not facial recognition. It's comparing image features against what exists on the internet. Famous faces in unusual contexts — casual photos, bad lighting, candid shots — often come back as "Human" or "Gentleman."

Richard Branson came back as "Jeffrey Epstein." Kevin Spacey came back as nothing. Ricardo Mangue Obama Nfubea — the Vice President of Equatorial Guinea — came back as nothing. I identified those manually, cross-referencing against the document index.

The machine found 15 named individuals. A face-blind human added 3 more. Together: 18 identified people in DOJ evidence photographs, plus 21 unknown faces awaiting identification.

The Marci Hamilton Screenshot

EFTA00037120 is a screenshot of Prof. Marci A. Hamilton's biography from childusa.org. She is the founder and CEO of CHILD USA, the leading expert on clergy sex abuse and child sex abuse statutes of limitations. She has testified before legislators in every state where significant statute of limitations reform has occurred.

Jeffrey Epstein had a screenshot of her bio in his files.

A man whose entire criminal enterprise depended on statutes of limitations was tracking the woman most actively working to eliminate them. Draw your own conclusions.

What This Cost

| Item | Cost |

|------|------|

| Face detection (MediaPipe) | Free (open source) |

| JP2 to JPEG conversion (Pillow) | Free (open source) |

| Face cropping | Free (Python) |

| Google Cloud Vision API (297 calls) | ~$0.45 |

| Meilisearch index (329,474 docs) | ~$15/month |

| Total infrastructure | ~$76/month |

| **Total for this analysis** | **$0.45** |

For 45 cents, a face-blind person identified 18 people in DOJ evidence photos and cross-referenced them against 329,474 indexed documents.

What I Need

Money.

There are 42,182 scanned evidence images. I analyzed the 279 that contain detectable faces. There are thousands more that contain handwritten notes, receipts, flight manifests, address book entries, and photographs that didn't trigger the face detector — group shots, partial faces, faces at angles.

With proper funding, I can:

- Run **OCR + document classification** on all 42,182 images (handwriting recognition, receipt parsing, manifest extraction)

- Deploy **dedicated facial recognition** (not just web matching) to identify the 21 unknowns and catch what Google Vision missed

- **Cross-reference every identified face** against flight logs, visitor logs, and the 329,474 text documents already indexed

- Build a **searchable photo evidence database** linked to the document index

- Process the **next DOJ batch** same-day when it drops (we indexed 329,474 documents while CNN was still writing headlines)

The entire Epstein evidence archive — 329,474 documents, 42,182 images, fully indexed, searchable, and cross-referenced — runs on $76 a month. Two people in Minnesota built this. The pipeline exists. The data exists. The methodology works.

I just need someone to pay for it.

**Contact:** [email protected]

**Search the files yourself:** [epstein.dugganusa.com](https://epstein.dugganusa.com)

---

*Every EFTA number is verifiable against the DOJ release at justice.gov/epstein. Every identification is documented. The machine found the faces. The face-blind guy found the names. That's the whole point.*

*Her name was Renee Nicole Good.*

*His name was Alex Jeffery Pretti.*