Gaille.

Gaille Amolong · Cebu, PH · remote

I ship production AI and security systems — and publish the proof.

SIEM for 170k+ users at Willed · 5,000+ provider health platform at Referrin · 20× CV pipeline speedup at BitWork · agent devtools on npm

Gaille Amolong
4,800+GitHub contributions in the last yeargithub.com/gael55x1,000+LeetCode problems solved · top 10% contest ratingleetcode.com/u/gael55x342npm packages published from open-source devtoolsnpmjs.com170k+users on the largest platform I engineer forwilled.com.au

every number links to its source

Selected work

Impact at scale

Problem, approach, measurable outcomes, and why it mattered for the business.

  • Enterprise SIEM for a 170k+ user platform

    Willed · Software Engineer, AI & Security Platform

    Problem
    A legal-tech platform serving 170,000+ users needed real detection visibility, SOC 2 readiness, and hardened authentication without slowing product delivery.
    Action
    Architected a custom enterprise SIEM with event-driven security pipelines, centralized alerting, and notification routing; built agentic security review and engineering workflows; enforced MFA and account lockout; led the Quasar-to-Nuxt.js frontend migration.
    Result
    Detection visibility across the platform, SOC 2-aligned controls and evidence, reduced account-takeover risk, and ~30% better frontend performance.
    Business impact
    Security posture that stands up to enterprise and compliance scrutiny while product velocity holds.
    • Nuxt.js
    • Node.js
    • PostgreSQL
    • SIEM
    • OAuth2/OIDC
    • CI/CD

  • Healthcare referral platform to controlled beta

    Referrin Health · Contract Software Engineer, Platform & AI Product

    Problem
    A healthcare referral SaaS had to move from development into controlled alpha/beta across a network of 5,000+ providers and case managers — with the rollout safety healthcare demands.
    Action
    Owned major platform engineering: stabilized referral, organization, provider, waitlist, and case-manager workflows; established staging/production environments, CI/CD, and E2E testing standards; built Smartforms AI modules for document handling and referral routing; hardened runtime secrets and encrypted configuration.
    Result
    Core workflows stable across the 5,000+ provider network, safer release cycles, and AI-assisted intake reducing manual referral processing.
    Business impact
    A product healthcare operators can adopt: compliant, predictable, and faster at moving patients through referrals.
    • React/Next.js
    • FastAPI
    • PostgreSQL
    • AWS
    • Docker
    • E2E testing

  • Badge Guru — computer vision pipeline for manufacturing

    BitWork Solutions · Lead Software Engineer

    Problem
    Manual emblem processing capped manufacturing throughput: every badge design took about 10 minutes of hand work before it could go to production.
    Action
    Built the computer vision R&D behind scanner automation: an OpenCV rendering pipeline with ArUco marker detection, glare reduction, and vectorization that turns a physical scan into production-ready assets.
    Result
    Emblem processing cut from 10 minutes to 30 seconds (~20x) — automated, repeatable, and no longer the bottleneck on the line.
    Business impact
    More units through the manufacturing line per shift without adding headcount.
    • Python
    • OpenCV
    • ArUco
    • FastAPI
    • AWS

  • 8BallDesigner.com — ecommerce serving thousands of customers

    BitWork Solutions · Lead Software Engineer

    Visit 8balldesigner.com
    Problem
    The store needed ecommerce reliable enough for a customer base in the thousands — checkout, shipping, and auth could not be weak links.
    Action
    Led engineering delivery across a lean frontend, backend, mobile, and DevOps team: architected Stripe/GoShippo ecommerce workflows, reworked auth with NextAuth + OAuth2/OIDC, and introduced blue-green deployments with rollback.
    Result
    Thousands of customers served with dependable checkout and shipment operations on a stabilized platform.
    Business impact
    An ecommerce store the business scales revenue on, with deploys that no longer risk downtime.
    • React/Next.js
    • React Native
    • Stripe
    • PostgreSQL
    • AWS

“Gaille quickly stood out as a brilliant full-stack developer. He is incredibly smart, hardworking, and dedicated, often going the extra mile to deliver high-quality work. He communicates openly, collaborates effectively, and uplifts those around him.”

Sheldon Arthur Sagrado · Full-stack Developer, Bitwork Solutions

Open source

Infrastructure I build in public

Shipped tools with commits, releases, benchmarks, and packages you can inspect — not weekend demos.

AI devtools

Snapline

Keeps AI-generated UI on the design system

0/30drift with hook on (vs 7–40% baseline)
240live agent sessions benchmarked
~126mshook overhead
What it is
A deterministic TSX scanner (TypeScript compiler API, no LLM) wired into Claude Code lifecycle hooks. It blocks agents mid-loop with exact repair contracts and gates completion until severe design-system drift is fixed.
The hard part
Proving it works: a reproducible benchmark of 240 live agent sessions (8 modes × 10 prompts × 3 attempts) — 0/30 drift with the hook on, versus 7–40% drift without it, at ~126ms overhead per hook call.
TypeScriptCompiler APIClaude Code hookspnpm monoreponpm: @usesnapline/cli

AI infrastructure

Grape

Context transport for AI coding agents

400+commits
14releases (v1.0.0 shipped)
31–52%context reduction, 3 fixtures
What it is
A repo-backed context continuity layer: dependency-tracked context artifacts compiled from the repository, per-session ledgers of what an agent has already seen, and MCP/CLI surfaces that send only what is new, changed, pinned, or stale.
The hard part
Diffing context honestly. Grape ships benchmarks showing 31–52% context reduction across three fixtures — published with the caveats, because a benchmark you can’t interrogate is marketing.
TypeScriptMCPCLInpmnpm: grape-context

AI education

Codemm

Local-first programming tutor with verified practice

430+commits
4languages sandboxed
Dockerisolated grading runtime
What it is
Chat a learning goal, get generated programming activities with tests, and run them in Docker sandboxes that grade real solutions — network-isolated, memory- and CPU-capped. Java, Python, C++, and SQL.
The hard part
Making arbitrary generated code safe to execute: per-activity sandbox isolation and a verification loop where the tests, not the LLM, decide whether your solution passes.
TypeScriptDockerLLM orchestrationSQLNow maintained under the iignaite org.

Quant research

LayeredMemoryTrader

Multi-agent LLM trading with layered memory

33★GitHub stars, 6 forks
3memory layers per agent
FAISSsemantic memory recall
What it is
Trading agents with human-style short/mid/long-term memory (FAISS-backed) that debate to consensus before acting, backtested against buy-and-hold with Sharpe and max-drawdown reporting.
The hard part
Getting agents to disagree productively: the debate mechanism weighs each memory layer’s recall differently, so a short-term signal can be overruled by long-term regime memory.
PythonFAISSLLM agentsBacktesting

More projects

Smaller builds that show range: mobile AI, edtech, and volunteer infrastructure.

BuhAI — Diabetes assistant in Bisaya

AI HEALTHCARE

BuhAI — Diabetes assistant in Bisaya

Personalized diabetes assistant for Filipinos: LSTM glucose forecasts (30/60 min) over CGM data, ChromaDB + RAG over logged meals, conversational guidance in Bisaya.

  • React Native
  • Python
  • LSTM
  • ChromaDB
  • RAG
  • Gemini
DSA Visualizer — Algorithms you can watch

EDTECH

DSA Visualizer — Algorithms you can watch

Step-through visualizations of core data structures and algorithms with interactive controls. 140+ commits, deployed and in active use.

  • Next.js
  • TypeScript
  • Framer Motion
MSHS official site — School infrastructure

INSTITUTIONAL

MSHS official site — School infrastructure

Volunteer-built official web app for Minglanilla Science High School: news portal, admissions info, reCAPTCHA-hardened auth. ~1.2k monthly users, zero downtime reported since launch.

  • Django
  • PostgreSQL
  • Docker
  • NGINX
Ren — Mood-adaptive support app

AI MOBILE

Ren — Mood-adaptive support app

React Native app with sentiment-aware flows and adaptive motivational content, backed by Flask + TensorFlow/NLTK. Designed, built, and shipped in one week.

  • React Native
  • Flask
  • TensorFlow
  • NLP
PhEASYCS — Physics tutoring bot

EDTECH / NLP

PhEASYCS — Physics tutoring bot

High-school project that placed 4th in the Division Science & Technology Fair: a physics Q&A bot built on TensorFlow NLP with TF-IDF features.

  • Python
  • TensorFlow
  • NLP

Writing

Engineering lessons, written down

I write about the engineering lessons you only learn after things break — security post-mortems, agent infrastructure, and quant systems. 375+ followers on Medium; the quant series is syndicated in InsiderFinance Wire.

All essays on Medium

About

Self-taught engineer from Cebu, shipping production software since I was 17 — the youngest engineer hired at BitWork Solutions, where I later led the team. Today I work across two continents: security platform engineering for Willed in Australia and healthcare platform work for Referrin Health in the US. I care about systems that hold up under scrutiny — SIEM pipelines, sandboxed code execution, agent tooling with published benchmarks — and I write about what breaks along the way.

Gaille presenting an LSTM neural network architecture slide to a full room
Walking a full room through LSTM architecture
Gaille speaking with a microphone at AI/DATA MiniConf
Youngest speaker at AI/DATA MiniConf, Cebu — Feb 2025
Gaille with fellow speakers holding certificates at the Accenture-hosted conference
With fellow speakers at the Accenture-hosted conference
PythonTypeScriptReact / Next.jsReact NativeFastAPIDjangoPostgreSQLAWSDockerTensorFlowOpenCVOAuth2/OIDC & SIEMCI/CD

Contact

Email is fastest. I typically reply within one business day.

gaille.amolong1@gmail.com

ResumeEmail