Systems→Data→Interface
Engineered as one thing
I build the parts other people would rather not: the data layer, the model layer, and the plumbing that keeps both honest in production. Four years of it, remote, shipping to teams who had to trust the output.
Four that describe the range: a signal platform, a protocol contribution, an analytics consolidation, and a hackathon win built on zero-knowledge proofs.
Automated a coaching business end to end: ManyChat capture, n8n and Make routing applications through qualification, booking, onboarding and follow-up — plus the operations behind a paid Skool community.
NLP sentiment modelling, technical indicators and whale-tracking combined into one signal surface — 35% better accuracy for 500+ active users.
Third-highest contributor on a DeFAI protocol doing roughly $30K ARR. Built the Hedera plugin layer that lets AI agents actually transact across venues.
EVM, Solana and SUI consolidated from three dashboards into one, cutting analysis time 45% and backing strategies that moved $2M+.
An on-chain game whose hard problem was never the game — it was proving each player is a distinct human. Solved with World ID zero-knowledge proofs.
Most crypto products break at the seams between the contract, the indexer and the interface. Owning all three is the whole advantage.
See the indexSolidity and Cairo in production, with the mechanism design to back it. Audited quadratic-funding work, staking and transfer flows, Sybil resistance via zero-knowledge proof-of-personhood.
Indexing, normalising and monitoring activity across EVM, Solana and SUI in one surface — then layering anomaly detection on top so the signal arrives before the loss does.
The plumbing between card networks and chains. PayPal, Stripe and Zelle into automated USDC settlement, running serverless so the cost curve stays flat.
End-to-end automation for operators: n8n, Make and Zapier wiring the back office together, ManyChat handling the front of house. Lead capture, qualification, booking, onboarding and follow-up running without anyone copying rows between tabs.
NLP sentiment modelling and ML pattern recognition wired into trading and ops surfaces — scored against outcomes, not demos.
What the system has to be true about, and what it does not.
Thinnest path to the risky part — contract, index or model first.
Typed adapters, error paths, observability. Happy paths are the easy half.
Into production behind monitoring, with the handover written down.
Bennett University
B.E. Computer Science & Engineering
2021 — 2025 · CGPA 9.3 / 10
Turing
AI trainer — labelled 1,000+ audio and voice samples for large-scale speech recognition and NLP training at 98%+ accuracy.
Freelance · 2026
Hackathons, open-source contributions, payment rails and analytics work — the full list rather than the highlight reel.
Built in 48 hours, won on stage
Shipped into protocols people actually run
TypeScript · Solidity · Cairo · Rust
Moving real money on-chain
EVM · Solana · SUI · Hedera
Where the AI half of the job lives
Influenced $2M+ in deployments across 5+ chains
Businesses that run without the founder in the loop
Contract and freelance work, India — Remote. Fastest route is email — include what you are building and where it is stuck.
aryantiwari7762@gmail.comThe profile on my résumé is my personal one, so its contribution graph looks quieter than the work actually is. Day to day I commit from my work profile, @harryhud, for my current company — several projects run there at once, which is why personal and open-source contributions have been lighter lately.
See @harryhud