GOKULKRISHNAN
VENKATESAN
Designing zero-knowledge protocols at the boundary of human intent and mathematical certainty.
A graduate researcher
writing protocols where trust becomes
proof.
Graduate researcher at the University of Illinois Springfield, working in applied cryptography under Dr. Goutham Reddy Alavalapati. Focus: zero-knowledge protocols that bind what users see to what their systems sign.
One paper currently under review at IEEE CNS 2026 — a publicly verifiable display-attestation protocol for cryptocurrency wallets (§ 02). 4.0 GPA. Teaching assistant, applied cryptography lab.
Outside the lab: top 2% globally on TryHackMe; CCNA-certified; maintainer of an open-source threat-intel CLI and an autonomous SOC agent (§ 03). Trajectory → PhD in applied cryptography.
VisiLock —
eliminating blind signing
VisiLock formalizes the missing property as display–transaction semantic alignment and constructs a publicly verifiable proof that it holds. A TEE module running in the Secure World — modeled on ARM TrustZone — reads the framebuffer directly from the display controller and computes a Poseidon commitment H_D over the rendered semantic regions. The transaction summary itself is rendered through a canonical UI overlay whose layout the wallet cannot alter, so H_D reflects what the display hardware is actually rendering rather than what the wallet claims to display.
A Groth16 zk-SNARK over the BN254 pairing-friendly curve then proves, without revealing any witness, that the extraction function Φ — mapping either object to a canonical tuple of (recipient, value, contract, function, parameters) — produces equal results on both sides: Φ(D) ≡ Φ(T). The R1CS circuit enforces that H_D and H_T are consistent with their preimages, that Φ has been applied correctly to each, and that every field matches component-wise. If a single field disagrees, the constraints become unsatisfiable and no valid proof can be constructed — the failure is detected at witness generation, not at verification.
Rather than parse arbitrary unbounded calldata in-circuit, VisiLock uses three fixed-size semantic-summary buckets: 48 bytes covers ETH and ERC-20 transfers and simple approvals; 100 bytes covers compact router-style swap summaries; 175 bytes covers larger bridge-intent and lending summaries (e.g., Across bridge intents, Aave V3 borrows). The 48-byte baseline circuit compiles to 228,514 R1CS constraints — text rendering through a depth-7 Poseidon Merkle font tree dominates at 52%, SHA-256 message hashing at 27%, comparators and routing at 20%. Crossing the Groth16 2¹⁸ → 2¹⁹ FFT-domain boundary at 175 bytes drives a 1.56× proof-time jump, which is precisely why the protocol routes between buckets rather than running one maximal circuit.
The on-chain verifier enforces five checks in sequence: chain-ID binding (cross-chain replay protection), canonical transaction commitment, TEE attestation against an allowlisted firmware- measurement registry with 5-minute quote freshness and a 24-hour validity window, Groth16 proof verification through Ethereum's BN254 precompiles, and quote authentication. A four-call batch mode aggregates per-call display and transaction commitments into a Poseidon binary Merkle tree and produces a single proof — reducing on-chain gas by 75% (1.08 M → 270 K) for typical DeFi sequences such as approve + swap.
The security argument is a reduction. Display Integrity follows from TEE isolation and Poseidon collision resistance. Semantic Binding reduces to Groth16 knowledge soundness under q-SDH and KEA, plus Poseidon collision resistance: any prover producing an accepting proof on (H_D, H_T) for which Φ(D) ≠ Φ(T) is converted, by Groth16's knowledge extractor, into a Poseidon collision — which occurs with negligible probability. The DisplayBind adversarial game generates 100 misaligned (D, T) pairs across recipient swaps, value inflation, selector mismatches, hidden-call injection, and parameter reorder, plus incident-driven scenarios reproducing the BadgerDAO and ByBit attack patterns. Every misalignment was rejected at witness generation.
The deployment cost of public verifiability over a TEE-only baseline is 2.4× — 192 K gas (signed assertion) versus 462 K gas (Groth16 proof). On L1 this is roughly $28 per transaction at 30 gwei; on Optimism it is under one cent, on Arbitrum and Base under ten. The deployment story is therefore L2-native: routine transactions stay on the cheaper TEE-only path, while high-value flows that need a publicly checkable alignment claim — bridges, large transfers, treasury moves — pay 2.4× for a property the wallet ecosystem has been missing for a decade.
Proof-Carrying Information Flow for multi-agent systems.
Replace content-level guardrails with information-flow cryptography. Every piece of data — user instructions, tool outputs, memory entries, retrieved documents — carries a label. Labels propagate as the agent reasons, recording exactly what influenced what in a provenance DAG that grows alongside execution.
Before any dangerous action, an admissibility gate evaluates whether the policy is satisfied across the action's entire ancestry — and admits the action only if a zero-knowledge proof attests that the lineage is policy-compliant. The verifier learns whether the proof is valid and nothing else: no internal documents, no memory contents, no reasoning traces ever leak.
Trust composes across agents. When one agent forwards output to another, it attaches a proof; the receiver verifies it and folds it into a recursive aggregate so a final verifier checks one constant-size proof regardless of chain length. A compromise at any agent breaks the chain at that boundary — and a tamper-evident commitment log enables retroactive blast-radius mapping when a source is later reclassified as malicious.
Beyond the proof —
instruments of defense.
IOC-Enrich
A professional-grade enrichment CLI. It detects an IOC's type — IPv4, IPv6, domain, URL, MD5, SHA1, SHA256, email — then concurrently queries VirusTotal, AbuseIPDB, Shodan, URLScan, ThreatFox and friends. Risk scores collapse into a single CRITICAL / HIGH / MEDIUM / LOW / CLEAN verdict, with rich color-coded terminal output and JSON / Markdown / CSV exports.
SOC-AI Agent
An end-to-end autonomous SOC analyst. It parses Sysmon XML, Windows Event Logs, firewall logs, .eml phishing, PCAP and raw text; extracts every IOC; concurrently enriches against seven threat-intel APIs; correlates against a SQLite history of past investigations; maps observed behavior against an 80-technique MITRE ATT&CK database via 30+ behavioral rules; and synthesizes a weighted verdict with a reasoning chain — delivered as streaming HTML and PDF incident reports through a React dashboard.