Research · Security · Cryptography
TITAN
Trustless Identity & Transaction Authentication Network
This started as a capstone project. It did not stay one. The deeper I dug into KYC verification pipelines, the more a single question kept surfacing: why does proving your identity require exposing your identity? Every bank, every exchange, every fintech app asks you to upload raw documents — passports, driver's licenses, utility bills — into systems you cannot audit. The entire trust model is backwards.
TITAN flips it. A verifier can confirm that an identity claim is valid without ever seeing the original document. That is not a theoretical convenience — it is a structural guarantee, enforced by mathematics.
"The ZK component was the hardest part. Circom's constraint system forces you to think about computation differently — every operation must be expressible as a rank-1 constraint system. You cannot hide behind abstractions."
Three Integrated Systems
Layer 1 — AI Document Forensics
Vision models trained to detect manipulated identity documents. Not just pixel-level tampering — the system catches font inconsistencies, metadata anomalies, and compression artifacts that survive casual editing. This is the first gate: if a document is forged, the pipeline rejects it before any cryptographic work begins.
Layer 2 — Zero-Knowledge Proofs (Circom)
The core innovation. Once a document passes forensics, its claims are encoded into a Circom circuit. The prover generates a ZK-SNARK that attests to the validity of the identity claim — age, nationality, residency — without revealing the underlying data. The verifier gets a boolean: valid or not. Nothing else leaks.
Layer 3 — GNN Anti-Money Laundering
Identity verification is only half the compliance picture. The GNN-based AML engine ingests transaction graphs in real-time, looking for structural patterns — layering, fan-out/fan-in, rapid cycling — that signal illicit flows. Graph neural networks are a natural fit here because the data is inherently relational.
System Architecture
Document Upload
|
v
+-----------------+
| AI Forensics |----> Reject (if tampered)
| (Vision Models) |
+-----------------+
|
v (clean document)
+---------------------+
| ZK Proof Generation |
| (Circom / R1CS) |
+---------------------+
|
v (proof + public signals)
+----------------------+
| On-chain Verification|
| (Smart Contract) |
+----------------------+
|
v (verified identity)
+-------------------+
| AML Engine (GNN) |
| Transaction Graph |
+-------------------+
|
v
+-------------------+
| Compliance Report |
+-------------------+