Hari Sankar

Built the production operating system for a startup-accelerator program that had outgrown spreadsheets and disconnected SaaS tools. Ness needed one system to find startups, onboard them, sell services, track cloud spend, and run every conversation without re-keying. I was a lead full-stack engineer from architecture through production.

KEY CONTRIBUTIONS
• Architected dual-face CRM and customer portal on React 19, Node.js/Express, PostgreSQL, Redis, and AWS ECS—5-role RBAC and SLA workflows replacing 5 disconnected operations tools.
• Built 15+ live n8n workflows ingesting email, WhatsApp, LinkedIn, and calendar into lead, contract, and AI-personalized outreach pipelines.
• Delivered Electron desktop reminders plus Hermes, an AI assistant on live program data.
• Production baseline: around 13,900 daily API requests at 99.68% success across 416,000+ monthly requests with dual-path alerting.

BITS became the single system of record from first lead through signed contract, cloud usage, and member support.

Built production AI leasing agents for a Canadian property manager losing after-hours inquiries and weekend viewing requests to voicemail—while staff repeated the same pricing, availability, and booking answers dozens of times daily. I owned the full-stack integration from CMS-driven property data through voice and chat agents to live booking flows.

KEY CONTRIBUTIONS
• Deployed per-property Retell AI voice and chat agents with 6 custom tools—search, availability, lead capture, rental applications, maintenance intake, and viewing booking—wired to Sanity CMS and a third-party scheduling API.
• Built the orchestration layer so prospects get pricing, submit applications, and book tours 24/7 without leasing staff on nights and weekends.
• Structured agent flows for real rental operations: property-specific knowledge, lead handoff, and booking confirmation rather than generic FAQ bots.

The system replaced repetitive leasing desk work with an always-on digital front line while keeping humans on high-intent viewings and applications.

Built Dropzon for Shopify and dropshipping operators who could see revenue in Shopify but not true profit after ad spend, COGS, and fees—forcing merchandising decisions onto spreadsheets updated days late. I was full-stack owner from architecture through merchant onboarding and live dashboards.

KEY CONTRIBUTIONS
• Designed a profit-analytics SaaS syncing Shopify orders and Meta Ads spend into per-SKU margin and ROAS views, answering which products actually make money after paid social.
• Implemented HMAC-verified Shopify webhooks, GDPR data-request/redact flows, competitor tracking, and a subscription academy on Next.js, Django REST, MongoDB, Redis, and AWS SQS/Lambda.
• Owned the full merchant lifecycle—OAuth, catalog sync, ad account linkage, dashboard rendering, and webhook replay safety for operators scaling Meta spend daily.

Dropzon turned delayed spreadsheet math into same-day merchandising decisions for operators running paid social at volume.

Built FittyMe for a consumer fitness product whose users quit within days, not because goals were wrong, but because manual calorie logging took minutes per meal. The bet was to photograph food once and let AI estimate nutrition instantly. I was the lead engineer across mobile, back-end, and production deployment to App Store and Google Play.

KEY CONTRIBUTIONS
• Shipped Flutter client (Riverpod, Supabase auth, Razorpay/PayPal, Apple Health sync) and FastAPI back end on AWS with multi-currency subscriptions, studio bookings, streaks, and personalized goal engines.
• Built async Vertex AI meal-recognition over WebSockets—job queue, retry logic, and confirmed meal logging—cutting nutrition capture from minutes to seconds.
• Owned recognition audit trails, deployment automation, and production hardening from photo capture through timeline update and analytics confirmation.

FittyMe removed the primary drop-off in calorie tracking and delivered a production consumer app end-to-end, not a demo model.

Built CxPlus AI for CX and product teams who saw customer feedback scattered across Trustpilot, Google Reviews, G2, and Google Play—but had no single view of sentiment, recurring topics, or what to fix first. Manual exports and slide decks took days and still failed to capture cross-channel patterns. I architected and delivered the full-stack platform as part of a funded client engagement.

KEY CONTRIBUTIONS
• Designed a modular microservices CX analytics platform aggregating reviews from Trustpilot, Google Reviews, G2, and Google Play with automated scraping, scheduled ingestion workflows, and real-time support-platform integrations.
• Built sentiment analysis and topic-extraction pipelines that turn raw reviews into ranked themes, representative complaints, and actionable insights on a unified dashboard—not vanity sentiment scores.
• Implemented alerting, scalable job orchestration, and service boundaries so new review sources and support tools could plug in without rewriting the core analytics path.

CxPlus replaced fragmented, channel-by-channel reading with a single operational view—surfacing what customers keep saying and where product and support should act next.

Built memgit—an open-source memory system for AI coding agents—after losing an entire workspace to disk failure and rebuilding from scattered notes. Developers and agent hosts had no durable, searchable, version-controlled way to persist decisions, gotchas, and project context across sessions. I was the sole product engineer from CLI through multichannel distribution.

KEY CONTRIBUTIONS
• Architected "git for AI memory," a Python CLI + MCP server storing typed mnemonics with BM25 search, checkpoints, supersession, and project scoping—dogfooded across 2,600+ operational memories in live engineering work.
• Shipped multichannel distribution on PyPI, npm (memgit-mcp), Homebrew, Chocolatey, VS Code Marketplace, and Claude Code plugin; built memgit.dev and an E2E-encrypted team-sync cloud layer on Azure Container Apps.
• Designed MCP-native hooks and tools for Claude Code, Cursor, and VS Code, so agents resume, recall, and save durable context without reloading entire instruction files every session.

memgit turned fragile chat history into versioned, agent-addressable engineering memory—local-first, MIT-licensed, and the context layer this workspace runs on daily.

ClutchX is a mobile marketplace for booking venues and experiences: party spaces, activities, and theme park tickets. Discovery, live availability, and payment happen in one app instead of across phone calls and messages.

I worked across the stack. The back end consists of six Java Spring Boot services (gateway, auth, booking, payment, notification, file) registered with Eureka, on MongoDB, with Redis streams carrying notification events, and Razorpay for payments. The Flutter app handles browsing, slot selection, and checkout, with a WebSocket layer for live availability and in-app chat. I also built parts of the React admin console used for venues, pricing, and community moderation.