Product engineer / systems thinker

ETHIANCHIU.

I build systems that turn messy operations into usable software.

Product Engineering / AI / Analytics / Systems

LEVEL -03

SELECTED SYSTEMS

Real builds at different stages of maturity.

01 / Logistics infrastructure
Deployed app2026

RouteFlow

Live worker tracking, multi-stop routing, ETA recalculation, and shareable public tracking views.

LIVE SYSTEM / ROUTE NETWORKPUBLIC VIEW + MANAGER VIEW
REALTIME
ETA
Problem

A manager coordinating several workers and deliveries needs route order, current location, and ETA in one place. Splitting those signals across maps, messages, and manual updates makes dispatch harder than it needs to be.

System

A Next.js application with a manager route planner, worker location flow, Supabase-backed realtime state, and tokenized read-only tracking pages for sharing a specific worker or route.

Next.jsTypeScriptSupabasePostgreSQLRealtimeMapLibreOSRM
Technical decisions
  • Use an organization-style manager view for multiple workers, trips, and ordered stops instead of treating tracking as a single-driver map.
  • Use Supabase Realtime so route and location changes can propagate without requiring manual page refreshes.
  • Use tokenized public tracking routes so a manager can share a read-only view without exposing the full manager workspace.
  • Use MapLibre + OSRM to preserve the project's no-paid-mapping-cost constraint instead of depending on a proprietary maps bill from day one.
Tradeoff

The open mapping stack keeps the demo inexpensive and portable, but it shifts more responsibility onto the application for routing reliability, service availability, and fallback behavior.

02 / Workflow intelligence
Active build2026

ApplyOS

A job-application workflow platform centered on document ingestion, structured application context, and assistive AI.

Problem

A job search quickly fragments into postings, resume versions, PDFs, application status, notes, and repeated decisions. The problem is not a lack of AI text generation; it is the lack of one coherent workflow around the application itself.

System

An in-progress application with a Python/FastAPI backend that treats resumes, job information, and application state as connected workflow inputs rather than isolated files.

PythonFastAPIPyMuPDFREST APIDocument parsing
Technical decisions
  • Start with a real backend and document pipeline instead of a chat-only prototype.
  • Treat resume and job documents as structured context that belongs to an application workflow.
  • Keep document processing behind API boundaries so the product UI can evolve independently of parsing and analysis logic.
  • Keep AI assistive: recommendations can reduce repetitive work, but the product should never fabricate qualifications or submit claims the user did not approve.
Tradeoff

Document and AI automation can save time, but application materials are high-context and representation-sensitive. The system therefore needs more explicit user control than a fully autonomous agent would provide.

DOCUMENT → CONTEXT → DECISIONACTIVE BUILD
01Job input
02Resume / PDF
03Document parsing
04Structured context
05User decision
03 / Financial intelligence
Deployed app2026

FINENGINE

A full-stack company-intelligence platform connecting live market telemetry, SEC fundamentals, peer benchmarking, filing evidence, and transparent valuation scenarios.

LIVE COMPANY DATATWELVE DATA + SEC EDGAR
Problem

Company research is usually split across market-data sites, spreadsheets, valuation work, watchlists, and SEC filings. That makes even a straightforward company review more fragmented than it should be.

System

A React/TypeScript and FastAPI application that combines Twelve Data market telemetry with SEC EDGAR/XBRL fundamentals, peer comparison, filing evidence retrieval, watchlists, and assumption-driven valuation scenarios.

PythonFastAPIReactTypeScriptTwelve DataSEC EDGAR / XBRLDocker
Technical decisions
  • Use SEC EDGAR/XBRL as the source of truth for company-reported fundamentals while Twelve Data handles market-price telemetry.
  • Normalize annual flow metrics separately from point-in-time balance-sheet facts so valuation ratios do not silently mix incompatible periods.
  • Use deterministic filing retrieval to return evidence passages and direct SEC links before adding any optional LLM layer.
  • Expose DCF assumptions directly instead of hiding valuation behind an opaque score or recommendation.
Tradeoff

Free and low-cost market-data sources vary in coverage, freshness, and rate limits. The application has to surface those boundaries rather than presenting every number as equally authoritative.

LEVEL -07

ARCHITECTURE

Different products. The same systems discipline underneath.

OPERATIONS
SYSTEM MODEL
DATA
INTERFACE
DECISION / ACTION
01

Model the real workflow

The system should reflect how work actually moves, including failure paths and handoffs.

02

Make state explicit

Important transitions should be visible, durable, and understandable instead of hidden in UI magic.

03

Integrate before replacing

Existing systems usually contain valuable data and process context. Replacement is not the default.

04

Use AI where ambiguity exists

Deterministic rules should stay deterministic. AI earns its place where interpretation adds value.

LEVEL -11

EXPERIMENTS / ARCHIVE

Smaller systems, prototypes, and technical studies.

LEVEL -18 / PROFILECREDIBILITY SIGNALS
01
AGENTIC AIAmazon AGI support

Evaluation / quality / failure analysis

02
TEAM9-person quality team

3,300+ AI tasks supported

03
RESEARCH1M+ healthcare records

ML / analytics pipelines

04
EDUCATIONMBA + BS Information Systems

Business analytics / computer science

THE PERSON BEHIND THE SYSTEMS

I work best where product, engineering, analytics, and operations overlap.

Professionally, I support Amazon AGI agentic-AI evaluation through a Keywords Studios engagement, working across model edge cases, failure analysis, quality workflows, and stakeholder feedback. I lead a 9-person technical quality team that has supported 3,300+ AI tasks, with work that reduced rework by 20%, improved first-pass accuracy to 95%, and increased productivity by 15%.

I also bring an MBA focused on Business Analytics from California State University San Marcos and a BS in Information Systems from San Diego State University with a Computer Science minor. My graduate research included ML and analytics pipelines across 1M+ healthcare records. I care about understanding the real process before choosing the technology: sometimes the answer is AI, and sometimes it is a better API boundary, validation rule, dashboard, or data model.