I build revenue-grade platforms — ERP & payments at the core, with AI tooling, realtime systems, and cloud infrastructure around them.
Leading 8–10 engineers on a multi-tenant Frappe/ERPNext platform · hands-on daily, from schema to Terraform to the LLM pipeline reviewing our PRs
| 📒 Particulars | Amount |
|---|---|
| 💸 Fee transactions verified, deduplicated, reconciled — HMAC-signed, idempotent, replay-safe | 100% |
| 🧩 Custom Frappe apps in production — thin overlays over one shared core, no forks | 25+ |
| 👥 Users served across 4–5 institutions, one multi-tenant core | 20–30k |
| 🔀 Payment gateways behind one pluggable, HMAC-verified router | 6 |
| 🏦 Bank payout rails, incl. ISO 20022 over GPG-encrypted SFTP | 8 |
| ⚡ Heaviest API latency, before → after (EXPLAIN + Redis) | 4–5s → <500ms |
| ☁️ Documented infrastructure cost reduction | −43% |
- 🧩 System design — multi-tenant architecture where correctness is the feature: idempotent webhooks, single-writer settlement, two-tier caching, graceful degradation by default
- 🤖 AI engineering — an LLM review pipeline that reads every PR with repo context before a human does; Claude Code & Copilot in the daily loop
- ⚙️ DevOps & platform — ~9k lines of Terraform (EKS, Multi-AZ RDS, IRSA, KMS), Helm + ArgoCD GitOps, zero plaintext secrets
- 💳 Payments engineering — high-volume fee collections through code I own: every transaction posted, verified, and reconcilable
| System | What it is |
|---|---|
| 💰 Fees & Finance Engine | 40+ DocTypes on ERPNext: payment plans, instalments, deferrals, discounts posting GL entries — re-architected as v2 with zero-downtime migration |
| 🏗️ Multi-Tenant ERP Core | 25+ custom Frappe apps, 250+ DocTypes, 500+ REST endpoints — overlay-app architecture, no forks, no monkey-patching |
| 🤖 LLM Review Pipeline | Every pull request read by an LLM, grounded in the changed code's context, before human review |
| ☁️ Cloud & GitOps Platform | Terraform-provisioned EKS, Helm + ArgoCD, DragonflyDB, External Secrets Operator — documented 43% cost cut |
| 🚌 Realtime Bus Tracking | Redis pub/sub → Socket.IO → React, with graceful WebSocket-to-polling degradation |
🔓 Public repos: frappe_toolkit · chatnext · chatpdf · looma
🔒 Client and institution names are kept private; the numbers are real. If you want the story behind any of them, write to me.
💬 Building on Frappe and hitting the same walls — payments, upgrades, infrastructure? Always glad to compare notes.