Sprite — full-stack engineer
Sprite.
17 years of full-stack and systems integration.Now building with AI — financial-institution automation, workflow optimization, adoption programs.
Full-stack and systems-integration engineer, 17 years of shipping products that stay in production. Seasoned across ITRI research, online gaming, crypto, and insurance. Beyond engineering itself, I've spent real time on the customer- and market-facing side — account manager, FAE, business development on contract work, and vendor coordination in my current role — leading teams of 10+ as an AM and an FAE. What sets me apart from most engineers isn't the stack: I switch language on the fly. Client, engineer, designer — each one gets the version that lands, and more of what I've shipped turned on that than on any framework. Now focused on AI in production — multi-agent systems, fintech apps, and browser automation at scale. Whatever the frontier is, I'm working at it.
What I bring
-
Multi-LLM, multi-agent orchestration — each model in its lane: cheaper tokens, fewer hallucinations.
By February 2026 I had the first version of this architecture done, built on two core design ideas:
- Spend tokens where they count — a cost/performance balance. Rather than throwing everything at the most expensive model, each agent stays in its lane. Groq runs a fast, cheap first pass of intent classification; only then does Gemini take over to receive and route the request, aligning with Claude on the requirement.
- Dynamic cross-verification — to kill hallucination. No model ever verifies its own work: requirements are settled by a Gemini–Claude debate and vote, then Claude handles core development while Gemini does the code review — the writer and the checker are always different models.
Underneath, Postgres holds persistent memory and Redis handles live coordination. The lab pattern ran for a few months before Sakana AI published the same concept as Fugu. The point was never stacking many models, or locking to one vendor — it was knowing exactly what each is for and putting cost where it belongs: cheap first-pass classification to Groq, Voyage for embeddings, Claude for the reasoning that has to be right.
-
Fleet-scale financial-institution automation — one control plane, three countries.
A Playwright fleet automates financial-institution portals across three countries — every institution isolated in its own browser session, coordinated across machines from a self-designed central control system, with non-intrusive, human-like interaction and login-error classification so one operator drives the whole fleet.
-
End-to-end product engineering.
WealthLens: from product philosophy to design system to Flutter implementation, solo.
Selected work
Financial-institution automation across three countries for a payments client — per-institution adapters that absorb portal churn, session isolation across parallel machines, and a self-designed control plane that load-balances and drives the whole fleet.
Read case study →Cross-platform personal-finance app for Taiwan's salaried workers. Task-driven, not another expense tracker.
Read case study →A fast-growing insurance brokerage, locked into its vendor: AI-assisted development at 5–10× my output — a year of internal systems shipped solo for a 700+ agent sales force, cutting admin processing time ~60%, with a RAG product-consultation service now underway.
Read case study →A personal lab where I ran AI agents like a software team — secretary, PM, engineer, QA — rebuilt three times. Same problems everyone hits with agents; the difference is I routed around them instead of grinding.
Read case study →Experience
Now
AI-assisted requirements analysis and internal tooling built at 5–10× traditional output, cutting admin processing time ~60%; vendor coordination. Technical lead of the IT unit — PM, RD, and the one driving the AI agents, in one seat.
Architected and led automated payment reconciliation on a distributed architecture, with central control coordinating multiple VM crawler nodes. Integrated non-intrusive techniques that precisely simulate real user behavior, keeping automated data collection within strict security and compliance boundaries — reliably supporting reconciliation for 50+ clients under the platform.
Side project
Contract work I take on when time allows — end-to-end hardware-software integration. Past projects include an iBeacon indoor-guide app for an art museum (native iOS/Android) and an RFID monitoring system for rail infrastructure.
The road here
- Solution-house engineering
Client-facing delivery across varied stacks and domains — the grounding in shipping to spec, on someone else's problem.
- ITRI · Research engineering
Engineering on ITRI research projects, including the i236 engineering data bank.
- Online gaming
FAE, platform-integration account management, and gaming platform systems — with a team of 10+ under me.
- Crypto · Market side
Crossed from build to distribution — grew the community, led a crypto token's initial fundraise, and ran investor sessions that translated the project's mechanics and the tech behind them into terms the market could act on.
Seventeen years, unafraid to cross into unfamiliar domains — reaching the core of each and delivering through systems integration.
Capabilities
AI orchestration & integration
multi-LLM routing, RAG (pgvector + Voyage embeddings), structured extraction, agent workflow design
Fleet-scale automation
Playwright, non-intrusive human-like interaction, multi-machine coordination, central control
Cross-platform product engineering
Flutter, web full-stack, design systems
Team & delivery
leads 4–12, vendor coordination, PMO, ISO 27001 exposure
Languages are an implementation detail — seventeen years from C to Rust means the next one is just syntax.
Toolbox: Python · TypeScript · Rust · C# · Dart/Flutter · Vue · Svelte · FastAPI · PostgreSQL · SQL Server · MySQL · Oracle · Redis · Docker · AWS · Cloudflare · Playwright · Claude API · Gemini · Voyage AI
How I work
-
Requirements first — at work.
Technology serves the requirement. Learn the domain language first — insurance, online gaming, crypto — then build.
-
Technology first — in the lab.
Some things deserve to be built for the technology itself. Minnow: an anonymous E2E-encrypted chat in Rust — no push, no likes, no analytics.
-
Lab first, production second.
New tech gets proven in my own lab before it ships at work. Hermes was the testbed for the AI-assisted workflow; putting it to work is what paid off.
Contact
Write me — a line about who you are and what you're building. I reply within 48 hours.
- Email: ·····
- GitHub: Spritech-Lab