Software architect & 0→1 product builder. Building AI-on-ERPNext — LazyChat (101-tool MCP server, open source) and LazyCode (AI coding agent for Frappe, private beta). Previously architected and shipped a Shopify-class omnichannel e-commerce stack end-to-end on ERPNext for a fashion D2C brand, from manufacturing through last-mile delivery, on one ledger.

Bangalore. Shipping in the open since 2013.

LinkedIn · Medium · X

I'm building lazychat-erpnext, an open-source AI assistant docked into the ERPNext desk. It's a 101-tool MCP server with bring-your-own LLM — any OpenAI-compatible or Anthropic key, the key stays in your browser. Mutations are Apply-gated; a vision-judge loop drives self-iteration. The Frappe community's first AI-native sidekick.

Alongside it, LazyCode.in (private beta) — an AI coding agent for Frappe and ERPNext apps. LazyChat helps users work in ERPNext. LazyCode helps developers build on it.

I architected and built a Shopify-class omnichannel e-commerce stack from zero for a fashion D2C brand — manufacturing to last-mile, reconciled to one ledger. Today, multiple ERPNext companies run on the same Frappe backbone in production. I own the end-to-end process design across P2P (Procure-to-Pay), O2C (Order-to-Cash), and R2R (Record-to-Report), and I designed and shipped the functional systems behind them: OMS, WMS, PIM, PLM, MES, marketplace listing and inventory sync, and product traceability with EAN and serialization.

The architecture separates Master Data, Transactions, and Financial R2R into three layers; B2B, D2C, and marketplace channels reconcile to one ledger. Forward and return flows include serial-number FIFO matching, isReturn invoice flagging, and automatic credit/debit notes. The webhook surface covers GRN inbound, the outward order lifecycle — create, complete, shipment-dispatch, deliver, cancel, partial-cancel — and channel returns. A Shopify D2C webhook stream handles order and customer events.

Live integrations run through a Java-based integration platform I architected — not a single middleware service, but a four-tier system: an ETL layer that ingests from ERPNext, marketplaces and channel-partner APIs; a workflow / data plane that runs orchestrated jobs with idempotency, retry and audit trail end-to-end; an API gateway with unified auth, rate-limit and observability fronting every downstream consumer; and a control-plane UI where operations teams inspect runs, replay events and manage connector state. The platform wires ERPNext to e-Waybill (Indian goods-movement compliance), Shiprocket + Cargofl (last-mile delivery aggregators), GSTR-1 (statutory returns automation), EAN barcoding for traceability and serialization, and five-bank H2H banking — ICICI, HDFC, AXIS, YES, Kotak Mahindra — via india-banking on Frappe. On the D2C ecosystem side, integrations include Shopify Plus, Gokwik (one-click checkout), and Return Prime (returns management). Full-ecosystem fluency with Increff covers WMS, OMS, CIMS, ICC, reports, and tech operations.

Stack: React + TypeScript + Tailwind + Vite + Storybook for the storefront; React Native and Flutter for companion mobile apps; Frappe/ERPNext (Python) with Java microservices, Node BFFs and Kafka on the backend; PostgreSQL, MariaDB, Redis and BigQuery for data. Infrastructure runs on GCP — Compute, VPC with load balancers, Cloud DNS, Cloud Storage — fronted by Nginx and Cloudflare, with GitHub Actions CI/CD across both the storefront and the integration platform. Multi-tenant deployments: multiple ERPNext companies share infra with isolated data.

The open-source pieces above are a slice. Behind the @agi-engg wall sit roughly fifty private services and apps I've architected and led: AI pricing-intelligence (ML-driven SKU pricing), AI customer chatbots, a Customer Data Platform (Java + Kafka + BigQuery, with a 360° view across web, app, marketplace and OMS), WMS extensions on top of Increff (bulk-picking, put-away), a 3D tech-pack tool for fashion design-to-BOM, warranty and return workflow apps on ERPNext, API gateways with unified auth, rate-limit and observability, shop-floor mobile apps for MES, an internal control-center dashboard that's the team's single pane of glass, and the corporate web properties — careers, policy portal, brand sites.

I run ERPNext in production across multiple companies — not a tutorial install, but a multi-tenant Frappe backbone serving live e-commerce traffic. Depth across all four layers:

Frontend. Custom Frappe Desk extensions; embedded React apps via iframe (the pattern lazychat-erpnext uses to dock into the Desk); DocType form scripts, dashboard charts, custom print formats, workspace customizations; client-side hooks for conditional fields, dynamic linking and validation; bench app bundles built and shipped via bench build + app_include_js / app_include_css.

Backend. Custom Frappe apps using the full DocType + Server Script + Hooks pattern. @frappe.whitelist API methods for external integration (lazychat alone exposes 101 of them). Document lifecycle hooks (before_save, on_submit, before_cancel, on_trash) driving downstream events. Background workers, scheduled jobs, queued events for long-running ETL. Multi-app dependency management across an evolving bench.

DevOps. frappe-docker production setup, multi-tenant bench architecture where multiple companies share infra with isolated data, GitHub Actions CI/CD for Frappe apps (test runners, app builds, deploy), version-upgrade workflows that survive ERPNext v13 → v14 → v15, automated DB backup / restore / migration, Nginx + Cloudflare in front of Frappe, Supervisor (or systemd on dev macOS) for process management, observability via Frappe's Error Log + custom log shippers.

Tech ops. Production support for live e-commerce traffic on the same backbone. Custom reports and dashboards for finance, ops, warehouse and growth teams. Role-based permissions and workflow approval flows. Cache and scheduler tuning, log rotation, queue depth monitoring. Daily reconciliation between ERPNext and external systems — Increff WMS, the five Indian banks, marketplace channels, payment gateways. Error monitoring, incident response, RCA discipline.

Started on Android in 2013, building consumer apps end-to-end. CS algorithm work in this era produced ShortestPath-CashFlow-Algorithm-Splitwise, an implementation of debt-minimisation across a graph of borrowers — still my most-starred OSS repo (15★, 10 forks). A handful of supporting libraries shipped alongside: extended layout managers, locked scroll views, and shared-preferences wrappers for Android teams to drop into existing apps.

The mobile era. React Native from scratch inside existing Android apps with auto-linking, OTA updates and Firebase wired in production. A Flutter clone of Inshorts (8★) shipped with native Chrome Tabs on Android and Safari View Controller on iOS for in-app web. The work here was as much about platform setup discipline — release pipelines, code-push flows, native bridges — as about the apps themselves.

Shifted up the stack — design systems at Figma-class scope, widgetized JSON-driven UIs, micro-frontend architectures, internal low-code tooling. Built component libraries that powered multiple downstream apps and a design platform layer that let product teams compose UIs from declarative JSON without writing React.

Built a Shopify-class omnichannel e-commerce stack from zero on ERPNext for a fashion D2C brand. Manufacturing through last-mile delivery, all reconciled to one ledger. P2P, O2C and R2R wired into one source of truth.

The Java integration platform I architected for this stack is a four-tier system, not a single middleware: an ETL layer pulling from ERPNext, marketplaces and channel-partner APIs; a workflow / data plane that runs orchestrated jobs with idempotency, retry and end-to-end audit trail; an API gateway with unified auth, rate-limit and observability fronting downstream consumers; and a control-plane UI where ops teams inspect runs, replay events, and manage connector state. The platform wires e-Waybill, Shiprocket + Cargofl, GSTR-1, Increff (WMS/OMS/CIMS/ICC), five Indian banks (ICICI · HDFC · AXIS · YES · Kotak Mahindra), Shopify Plus, Gokwik, and Return Prime.

Alongside the integration platform: a Java + Kafka + BigQuery Customer Data Platform unifying a 360° view across web, app, marketplace and OMS. ETL data pipelines (cron + GCS + OneDrive sync) for finance and ops reporting. Internal control-center dashboards giving the team a single pane of glass across the stack. Microservices for payments, auth, user-data and material catalogue. WMS extensions on top of Increff (bulk-picking, put-away) for shop-floor speed. Shop-floor mobile apps for MES.

DevOps and infrastructure: GCP-native — Compute, VPC with load balancers, Cloud DNS, Cloud Storage, AI Studio — fronted by Nginx and Cloudflare. GitHub Actions CI/CD across storefront, mobile apps and the integration platform. Multi-tenant deployments where multiple ERPNext companies run on the same Frappe backbone with isolated data and shared infra. Production today.

Building AI-native tooling on top of the ERPNext stack I run. LazyChat — open-source AI assistant docked into the ERPNext desk, 101-tool MCP server, bring-your-own LLM (any OpenAI-compatible or Anthropic key stays in your browser), Apply-gated mutations, vision-judge self-iteration. LazyCode.in — AI coding agent for Frappe and ERPNext apps (private beta). The first AI-native sidekicks for the Frappe ecosystem.

If you run an ERPNext shop and want AI on top, start with lazychat-erpnext.

If you're building a 0→1 product and need a software architect or technical co-founder, LinkedIn or DMs open on X.