Portfolio index

Selected systems & case studies

Eight production-minded projects spanning Sui Web3 (agent treasuries & social dApps), fraud detection, document RAG, real estate SaaS, dual-engine TTS, and AI-assisted CLI tooling. Each write-up covers problem framing, architecture choices, and engineering tradeoffs.

Web3 · Sui

W1 · AI billing · Move

SEAL

Sui Move TypeScript Next.js x402

Full case study GitHub

SEAL is a non-custodial billing layer for AI teams on Sui. Deposit once, set hard spend caps enforced by Move smart contracts, and route API calls to Groq, OpenAI, Anthropic, or x402-compatible services through a single API key.

Architecture splits into three layers: a TypeScript gateway (auth, provider routing, velocity tracking, batch settlement), on-chain contracts (deposit/withdraw, caps, authorize_call, pause), and a Next.js dashboard for wallet connect, treasury setup PTBs, key management, and anomaly alerts.

W2 · Agent treasuries

Otter

Sui Move Node/TS gateway TS + Python SDK

Full case study GitHub

Otter (Autonomous Agent Treasury Protocol) gives every AI agent its own on-chain treasury — a Sui object that holds Coin<SUI>, enforces spending policies, and can be reclaimed atomically. No central custodian; no opaque credit-card billing for agents.

Master treasuries spawn child agents with sub-budgets; daily/monthly/single-spend caps and provider allowlists are enforced in Move. The gateway settles each inference via authorize_agent_call and exposes TypeScript and Python SDKs.

W3 · Social dApp

DSpace

React 19 Sui dapp-kit Firebase Vite

Full case study GitHub Live site

DSpace (SuiPod Chat) is a hybrid decentralized collaboration platform: Sui wallet identity and on-chain pod membership, plus Firebase Realtime Database for instant messaging.

Users create pods, add members by wallet address, chat in real time, and send SUI directly to pod members without leaving the UI — bridging social product UX with on-chain payments via @mysten/dapp-kit.

AI · Fullstack · Tools

01 · AI · Fintech risk

Asguard

High-volume fintech environments need fraud decisions that are fast enough for live payment flows and explainable enough for compliance. Asguard scores transactions with layered rules, enriches them with AI risk assessment, and emits structured alerts with natural-language summaries.

Clear-cut decisions stay on a low-latency path; ambiguous events escalate for deeper model analysis with investigator context in-band.

Go Gin Firebase LLM PostgreSQL

Full case study Live site

02 · RAG · Document intelligence

Azeru

Enterprise RAG platform that ingests PDFs, chunks them for embedding quality, stores vectors for semantic retrieval, and answers questions with source-aware conversational responses.

Backend processing in Go for throughput; operator UI in Next.js. Inference routes through high-speed LLM providers so retrieval + generation stays interactive.

Go Next.js Groq AI Vector search Docker

Full case study Source

03 · SaaS · Real estate

Hauser

Real estate marketplace for Nigerian property professionals — agents, landlords, and tenants. Role-aware onboarding lands each persona in the right workflow without manual admin setup.

Next.js + Firebase (Firestore, Cloud Functions): dynamic listings, real-time updates, image optimization, and click analytics for premium tiers.

Next.js Firebase Firestore Cloud Functions

Full case study Live site

04 · AI audio · TTS

Koe

Next.js Azure TTS WebSocket Docker · Piper

Full case study

Dual-engine TTS studio: Microsoft Azure Neural voices for premium cloud synthesis, and offline Piper TTS via Dockerized local inference when data should not leave the machine.

Audio streams in real time over WebSockets. Engine selection is modular so operators can switch at runtime based on latency, cost, or privacy — useful for content studios, accessibility tools, and air-gapped environments.

05 · CLI · Devtools

Vpipe

Go Groq OpenAI Anthropic

Full case study

Go CLI that pipes compiler and runtime errors into multi-provider LLM APIs and returns actionable fix suggestions without a browser workflow.

Before any external call, auto-sanitization masks 20+ sensitive patterns. Smart truncation respects provider token limits — a security-aware take on AI-assisted debugging.

Want the architecture walkthrough live?

Happy to discuss tradeoffs, stack choices, and what I’d harden next in a technical interview.

Email Kennedy More ways to connect