TrustFlow
Milestone escrow powered by AI
Created on 5th September 2026
•
TrustFlow
Milestone escrow powered by AI
The problem TrustFlow solves
In freelance and service-based work, trust frequently breaks down before any payment actually occurs because informal communication over chat, email, or calls leaves payment conditions, deliverables, and revision rules open to conflicting interpretations. This creates a stalemate where clients risk losing their money by paying upfront for incomplete or substandard work, while contractors risk working without compensation if payment is withheld. Without a shared, objective source of truth, vague agreements inevitably result in payment friction, stalled projects, and contentious disputes. TrustFlow resolves this dilemma by bridging casual conversational terms into deterministic, milestone-structured agreements backed by non-custodial smart contracts, protecting both parties with clear review windows, objective evidence requirements, and controlled fund releases.
Challenges we ran into
AI Hallucinations in Contract Terms: The language model initially assumed missing details, such as fabricating wallet addresses or guessing review deadlines. We resolved this by constraining the AI strictly to data structuring, adding a deterministic Zod validation layer, and introducing an explicit "Help us fill in the gaps" step for mandatory human input. Upstream AI Outages & Timeouts: Live API calls to the LLM backend occasionally timed out or returned unexpected formats. We fixed this by saving raw terms as local drafts first and adding a fallback manual entry path that enforces the same strict validation rules. Balancing Privacy with On-Chain Integrity: Writing plain-text contracts and evidence to the blockchain would expose private business data and increase costs. We implemented a hybrid architecture where sensitive records remain off-chain in PostgreSQL, while only an immutable SHA-256 agreement hash and coordination anchors interact with Sui. Escrow Math & State Safety in Move: Ensuring locked funds could never be double-spent across multi-milestone lifecycles or disputes required rigid accounting. We built and verified 18 Move unit tests to strictly enforce single-release invariant checks and freeze disputed milestones until resolved. Mobile Key Management: Exposing private keys or burdening users with seed phrases undermines mobile security and UX. We decoupled custody by leveraging native Google sign-in with Enoki nonces, generating unsigned transaction builders locally, and never storing keys on the backend.
Tracks Applied (1)
PAYMENTS & STABLECOINS
Sui Foundation
Technologies used