Purpose

Primarily delivering personalized ecommerce product recommendations across key channels.

UI / UX Demo - Dashboard

Dashboard featuring analytics, insights, and other management like templating. Built using Vue/Nuxt with basic support of Buefy (based on Bulma) for consistency, and lean custom styling for a clean purposed UX. Use of simple visualizations such as line / momentum charts, funnels using a mix of open-source charting libraries including chartjs, echarts, d3.

Architecture Overview

Resource-scarce Engineering

Light infrastructure and custom deployments were always favored over enterprise / managed / resource-intesive infra.

This 0-expense model always dissolved to guerilla DevOps.

• Custom event pipeline in Node.js
• Use of MongoDB as data store
• PM2 and supervisor for process management
• Miniscule downtime shell push-deployments and minimal use of gitlab CI
• Redis for PubSub, queueing, and inter service communication
• Free SaaS for end-service monitoring
• Celery for scheduling jobs, and flower for manual disaster management
• Shell scripts for forensics on nginx logs

All self-hosted and operating on a single EC2 with 2GB RAM leaving enough runtime breathing room.

Concurrency issues were anticipated and systems were built as borderline microservice architecture and for readiness of scalability into a distributed architecture.

Near EOL: Data had scaled to about 1GB over months with avg 3 live stores, batch modeling was clocking several minutes, system was capable of being juiced out further with data pruning and compression at injest. Beyond that, doubling the memory capacity was estimated to send the mid-traffic (250k visitors / mo.) ecommerce customer ceiling to about ~20, theoretically sufficient for beta launch.

Custom Libraries

• Auto Scaffolder [published:npm:deprecated(2021)]
• connect.js

Notable Custom Internal Modules

• ISC Lib - Thin PubSub client loosely coupled with redis
• Event Service - Node.js implementation of PredictionIO schemas

Early Work

• Technically assessing competitors by time-in-market
• Building A/B testing suite for efficacy of machine learning models against various data volumes
• Reviewing hierarchichal topic modeling for content personalization

Wordpress / Woocommerce Integration

WooCommerce integration allowed 0-config tracking on woocommerce stores. It was refined over the life of the product with feedback across different scenarios while in production.

Open Sourced @ Github: personide/personide-for-woocommerce

Shopiy App Walkthrough

State

Status: Retired

Last release: first alpha

Org Challanges: Cash scarcity -> foundership sustenance gap -> inhibited outreach -> Lack of GTM -> pollutant 3rd actors -> fog-of-communication -> feature creep -> lost sovereignty

Tech Resolve (pending resources):

• Establishing tests coverage
• Migration to planned relational data warehouse from document DB
• Trial of pentaho mondrian for ETL
• Encoding event data into custom protocol
• Migrating analytics to composable functions using custom parser

Unexecuted Pivot - Phase 1: Splitting product into two -> API-only and partnered Recommendation Delivery Platform

Cause of retirement: Derisking on part of engineering founder (myself)

Screens