Upload a resume, salary slip, offer letter, relieving letter, Aadhaar card, PAN card, or bank document — as a PDF or a photo. 12 real forensic checks run against your file live: document metadata, OCR confidence, field arithmetic, error level analysis, font consistency, photo splice/face-region consistency, a duplicate-identity scan, PAN and Aadhaar (UIDAI Verhoeff checksum) validation, bank detail validation, a resume authenticity ensemble, and a final AI-reasoned verdict. No canned demo numbers.
Real backend. Your file never leaves this scan.
Every check below runs against your actual file — highest-signal, lowest-effort forgeries first, AI reasoning last to synthesize everything.
Reads PDF/EXIF metadata for editing-software fingerprints (Photoshop, Canva, GIMP…) and creation-vs-modification date mismatches.
Confirms the document's text was reliably extracted — native PDF text layer first, real Tesseract OCR confidence as a fallback signal.
Cross-checks gross − deductions = net, and that joining / relieving dates are in the right order relative to today.
Recompresses the document and diffs it against the original — pasted or re-edited regions recompress differently and light up.
Flags an unusually high number of distinct fonts/sizes in the body text — a common side-effect of manual edits.
Recompresses the image and looks for a spatially contiguous region with elevated recompression error — the signature of a pasted/composited region (a swapped face, a copy-pasted photo). A regional-consistency heuristic, not a neural deepfake classifier.
Blocks by phone/email/name key against every prior submission, then an ensemble of a Random Forest and a small neural net scores name variants, DOB, device signal, IP, and bank account — matches cluster into an identity graph.
Validates the PAN format (5 letters + 4 digits + 1 letter) and its 4th-character holder-type code, and flags multiple conflicting PAN numbers in the same document.
Verifies any Aadhaar-shaped 12-digit number against UIDAI's own Verhoeff checksum algorithm — a random or mistyped number fails this with near-certainty.
Checks IFSC code format, MICR code length, and flags account numbers that look typed-to-look-real (repeated or sequential digits) rather than issued.
Weighs employer/domain cross-referencing, file-metadata forensics, embedding similarity against every resume on file, and an AI-text classifier into one ensemble verdict — no single signal decides alone.
An LLM reads the document text plus every signal above and renders a final risk verdict.
Upload it and let the agent run six real checks in the time it takes to read this sentence.