Inspiration
Crypto wallets are intimidating. Private keys get lost, stolen, or exposed — and the UX is terrible for newcomers. We wanted to prove that enterprise-grade security and dead-simple usability aren't mutually exclusive. AWS KMS and conversational AI made that possible.
What it does
HASHFLOW lets users send HBAR, check balances, and view transaction history using plain English. Type "Send 50 HBAR to 0.0.12345" and it executes securely on Hedera Testnet. The private key never exists outside AWS KMS hardware — not in memory, not in env vars, nowhere.
How we built it
Built with Next.js 14, Hedera SDK, and AWS KMS asymmetric ECC_SECG_P256K1 signing. The Hedera SDK's setOperatorWith() method accepts a custom signer — we plugged AWS KMS directly in. A locally running Gemma 3 4B model via LM Studio handles natural language intent detection. Every signing event is logged to AWS CloudTrail automatically.
Challenges we ran into
AWS KMS returns ECDSA signatures in ASN1 DER format but Hedera expects raw 64-byte r+s encoding — we built a custom parser to bridge this. Hedera's strict transaction timestamp validation caused silent rejections until we fixed it using TransactionId.withValidStart() set 30 seconds in the past.
Accomplishments that we're proud of
The private key genuinely never exists outside AWS hardware — not even for a millisecond. We also built a fully local AI pipeline where zero user data leaves the machine. Together these make HASHFLOW one of the most secure and private wallet implementations on Hedera.
What we learned
Learned the full AWS KMS asymmetric signing flow, ASN1 DER signature parsing, Hedera's custom signer architecture, and how to make a local LLM reliably detect transaction intent without tool calling. Security and simplicity can absolutely coexist.
What's next for HASHFLOW
HTS token support, scheduled recurring transactions, multi-account management, HashPack wallet integration, and on-chain AI queries — "How much did I spend last week?" answered directly from real transaction history.
Built With
- amazon-web-services
- hedera
- kms
- lmstudio
- nextjs
- tailwind
- typescript