I build line-of-business software — the CRM, ERP, hospital and HR systems a company's daily operations depend on. At Hybrid Interactive I have shipped eight production products across healthcare, hospitality, construction, calibration, sales and HR.
The hard parts of this work are rarely the framework. They are the tenancy model, the audit trail, the state machine that decides what a document is allowed to become next, and whether the person who picks the code up after me can still read it. My largest system is a lead-to-cash CRM for a UAE medical-equipment calibration laboratory: 18 business modules, multi-branch, audit-backed, built to contractual SLA targets. I wrote 98% of both repositories.
On the AI side I integrate LLM features into real product surfaces — Pydantic AI for typed agent behaviour, RAG pipelines over client data, and direct API calls where a full agent framework would be overkill. I work agentically day to day with Claude Code, Codex and Copilot, with deliberate judgement about where a model speeds up real progress and where it just adds noise.
Client work is what I ship. These are what my team builds on.
A portable folder you copy into a new backend on day one. 30 concern guides covering permissions, migrations, caching, tenancy, timezones, money handling and state machines — plus a checker that fails the commit when a doc goes stale. Every rule in it was paid for once already, by a bug that shipped. |
Built and documented on my own. 143 files: BFF cookie auth so tokens never reach JavaScript, RBAC gating navigation and components, TanStack Query, Vitest and Playwright. Documented in 17 numbered rule guides so the reasoning survives me. |
Also: FastAPI Backend Template (co-designed — routes hold no business logic, services own the transaction boundary, data access never commits) and the company's public Engineering Standards hub.
The twenty decisions that get expensive in month six →
Tenancy, audit and money are cheap to decide before the first module and painful to change once there is data.
| Area | Focus |
|---|---|
| Backend | FastAPI · async SQLAlchemy 2.0 · PostgreSQL · Redis · Celery · Alembic |
| Frontend | Next.js App Router · React 19 · TypeScript · TanStack Query · Tailwind |
| Architecture | Modular monoliths · multi-tenancy · RBAC · append-only audit trails |
| Reliability | Background workers · caching · Prometheus · structured logging |
| AI in product | Pydantic AI · RAG pipelines · LLM APIs in real workflows |
| Practice | Written standards · architecture decision records · docs that carry a date |
| Project | What it is | Stack |
|---|---|---|
| Nadisa Solutions | Live career-guidance platform for chemistry professionals. Built solo. | Next.js 16 · React 19 · Firebase |
| Code Eval AI | AI-assisted code and interface review workspace with structured scoring. | React · Node · PostgreSQL · Gemini |
| ClickTalk | Realtime chat with presence and secure auth. | MERN · Socket.IO · Zustand |
| BookStore | Catalogue and inventory interface. | Next.js · PostgreSQL · Drizzle |
The client work is described — without client names — at avi9611.github.io.
Build for the person debugging this at 2am, not the person reading it today.
Open to conversations about backend architecture and product delivery.