I'm a platform product manager with 23+ years spent at the seam where deep technology meets real-world use — distributed systems, data infrastructure, AI platforms, and applied cryptography. Along the way I've built and shipped products across cloud infrastructure, developer platforms, privacy and cryptography, and verifiable AI & financial systems.

This repo is a map of the things I build to think — open-source projects, prototypes, and demos. Most of it is me chasing a question end to end: what does it actually take to make this work? Browse by theme below, or jump straight to whatever caught your eye.

Most of this is public and meant to be runnable. That said, some pieces — especially the older ones — may have gone stale or stopped working as dependencies, platforms, and networks moved on. Browse all repositories →

yutha A framework-agnostic control plane for groups of AI agents — giving agents from any framework (LangGraph, CrewAI, …) shared identity, attenuated capabilities, signed receipts, declarative constitutions, and an optional on-chain verification layer. Multi-agent systems work in demos and break in production; Yutha is the missing substrate. Docs at yutha.ai.

Prototypes exploring how cryptography (Seal MPC, Confidential Transfers) and controlled execution let money and markets move on public chains without leaking amounts — and let autonomous agents transact without a trusted middleman.

seal-agentic-finance Agentic finance on Sui + Seal MPC: AI agents compete with sealed, deposit-backed bids that stay hidden until a deadline, reveal on-chain all at once, and bind the winner to their terms. One auction engine, two markets — best-execution RFQ and lending.

seal-programmable-wallet Seal MPC as the authorization layer for agentic commerce: users grant agents scoped, programmable spending mandates, and an MPC committee — not the agent — decides whether each payment is allowed. A compromised agent can't lift its own limits.

confidential-autonomous-treasury "Helm" — an AI agent runs a company's stablecoin treasury on Sui: sweeping idle funds, allocating across vetted venues, replenishing a buffer, and paying vendors under hard limits it can't break. Every action is authorized outside the agent by Seal MPC + on-chain policy, balances and amounts stay encrypted via Confidential Transfers, and a risk committee gets read-only oversight.

confidential-payroll-sui "Stipend" — confidential payroll on Sui. Salaries are encrypted on-chain, each employee sees only their own pay, a finance lead can audit read-only, and employees can prove their income to a lender without revealing any key. Built on Confidential Transfers.

confidential-b2b-settlement "Tessera" — a permissioned consortium settling invoices in a confidential stablecoin, with KYC-gated membership, payment channels, disputes, and regulator oversight turned into scoped, time-boxed, revocable Seal MPC mandates instead of one permanent skeleton key.

treasury-rfq-demo A clickable prototype of an institutional treasury workflow on Sui Spheres — scoped multi-dealer RFQs, sealed quotes, loser-blindness, settlement that bridges to public Sui, repo lending, and cause-based regulator reveal. Makes the multi-party privacy story tangible end to end.

sui-spheres-supplychain-finance A Sui Spheres prototype of supply-chain financing: a commerce group underwrites its merchants against invoices, inventory, and orders using its own private data, while an outside investor fund supplies the capital — funding against redacted, verified positions without ever seeing the merchant's private information.

sui-spheres-interbank-settlement A demo of interbank settlement — within and across countries — built on Sui Spheres.

Sui Spheres prototypes where agents across parties — several run by live AI — compete or collaborate under rules the infrastructure enforces, each acting on a private view no one else (not even the operator) can see.

sui-spheres-energy-markets An energy flexibility market where rival energy companies — two run by reasoning AI agents — bid to keep a power grid balanced, get cleared cheapest-first, deliver, and get paid in real time. Everyone sees only their own offers, no one can over-promise, and deals are binding.

sui-spheres-intraday-collateral Firms' AI agents answer intraday margin calls under a hard deadline, pledging from private inventory the operator can't see and swapping assets atomically — racing across two settlement rails: a slow custodian versus tokenized collateral bridged instantly from public Sui.

sui-stack-crm A small-team, Airtable-like CRM where the team — not a vendor — owns the data. Records are encrypted in the browser with Seal, stored as content-addressed blobs on Walrus, and gated by an on-chain ACL on Sui, so no operator can read them. A concrete take on what a privacy-first SaaS alternative looks like on the Sui stack.

seal-mpc-ops A REPL coordinator for running Seal MPC committees on Sui: it folds the full lifecycle — DKG ceremony, on-chain register/propose, aggregator deployment, and member rotation — into one coordinator-driven tool, with --json output so an AI agent can orchestrate committees end to end.

multi-agent-verifiable-research A multi-agent research lab that turns a free-text question into a citation-grounded report. A planner, parallel readers, a critic, and a synthesizer collaborate through shared memory — every source is content-addressed on Walrus and every claim is a clickable, verifiable citation.

sui-trader-agent A complete sample agent that uses MemWal for durable memory and Claude for reasoning, scoped to a paper-trading persona. A live inspector shows what the agent stores, retrieves, and how that memory shapes its next move — in real time.

Tools built on Sui's post-JSON-RPC data stack (gRPC, GraphQL RPC, Archival Service) — several exploring how AI agents can query a chain without a multi-week SDK integration.

sui-mcp-server A local-first MCP server that lets any agent — Claude Desktop, Claude Code, a custom loop — query Sui in plain English across gRPC, GraphQL, and the Archival Service. Handles transport selection, retention boundaries, and LLM-friendly response shaping so the agent just answers the question.

sui-data-stack-explorer An AI-native Sui explorer that routes across all three transports and shows you why in a routing-trace panel. A Claude-powered assistant picks the right transport per question, with live checkpoint streaming and object time-travel.

sui-explorer-grpc-graphql A read-only reference-architecture explorer built end to end on the new data stack — a clean starting point showing how each API maps to real explorer features as JSON-RPC sunsets.

Grouped by topic rather than timeline: provisioning and hardening Databricks on AWS/Azure, and hands-on stream processing with Spark, Storm, Kafka, and Flume. Some of these go back a while — I try to keep the list current regardless.