Open to remote or relocation

Senior Software &
Platform Engineer

Backend, infrastructure and distributed systems.

15+ years building and running web products end to end — the database, the servers, and the UI on top of them. I lead engineering at MeNEW, where I cut the infrastructure bill nearly in half while the platform grew 40%.

Numbers I can defend

Every figure below comes from production work I did myself, and I am happy to walk through how each one was measured.

−47%
infrastructure cost

$1,500 → $800 a month, while active venues grew from ~500 to ~700

10×
faster menu load

GraphQL responses cached in Redis with stale-while-revalidate and stampede protection

−80%
database writes

400 → 80 commits a second at peak, by writing only on real change

120 fps
menu editor, up from 20

Same machine, same features — memoized board rendering

70%+
CDN cache hit rate

Pre-warmed image variants served from a separate optimized bucket

30s
3D scene generation, down from 5 min

Same RTX-5090 hardware, exported assets half the size

Where I have worked

Fifteen years across four companies in Iran, the UK, the United States and Canada — the last five entirely remote.

Co-Founder → Tech Lead & DevOps

2017 – 2020 · 2025 – Present

MeNEW — Digital menu platform · Remote Current

I co-founded MeNEW in 2017 and built its first version, then came back to lead it technically. It is now Iran's largest digital menu platform: ~700 active venues on a multi-tenant Django / Next.js SaaS, handling around 200 requests a second at peak and 100,000 menu views a day. I lead four engineers and own the architecture, the infrastructure and the cost of running it.

  • Halved the infrastructure bill and spent the savings on redundancy: an active and a passive server with PostgreSQL replication, plus an independent backup stack on its own domain, CDN and server.
  • Designed the GraphQL response-caching layer on Redis and mentored the team through it — the backend went from 8 nodes at ~300% CPU to 2 nodes at ~200%.
  • Re-architected subscription and billing, splitting invoice bookkeeping from provisioning so paid invoices stopped silently failing to deliver.
  • Resolved a PgBouncer connection-exhaustion outage with concurrency limits and refusal caching.
Django 5.2Graphene GraphQLCeleryPostgreSQLTimescaleDBRedisNext.js 15React 19

Technical Consultant

Dec 2025 – Present

Imersual Ltd — Remote · United Kingdom Current

AI-powered immersive and creative platforms: 3D environment generation, storyboard automation and cinematic pre-production.

  • Lead development of BeeStoon AI, which turns a single image into an explorable 3D environment using Gaussian Splatting and mesh reconstruction.
  • Cut scene generation from about five minutes to thirty seconds on the same hardware, with exported assets roughly half the size.
  • Configured GPU infrastructure on Linux with CUDA, PyTorch, Mamba and Docker for large-scale 3D inference.
  • Built Storyboard Studio: drag-and-drop scene composition, animatic workflows, shot configuration and production breakdowns.
Gaussian SplattingPyTorchCUDAComfyUINext.jsDocker

Lead Software Engineer

Mar 2023 – May 2025

Airdec Inc. / OZure — Remote · Virginia, USA

Core backend and frontend systems for immersive AI-driven platforms and internal business applications.

  • Product owner for three of the company's applications, from requirements through to release.
  • Took the immersive scenes from under 10 fps to a steady 50–60 fps through draw-call reduction, asset compression and mesh optimization.
  • Built an NLP and ETL pipeline over technical drawings to auto-extract project name, address, architect and engineer.
  • Ran a Proxmox host with VMs for a Qdrant vector database, Helix Perforce Core, NextCloud, staging, backups and GitLab.
Django DRFPlayCanvasThree.jsOpenAIQdrantProxmox

Senior Frontend Developer

Jan 2021 – Jan 2023

Reev Tech Inc. — Remote · Toronto, Canada

High-performance web and plugin-based applications across several verticals.

  • Launched five production Vue.js and vanilla JavaScript applications.
  • Wrote an optimized ES6 module system with a custom HTTP abstraction layer.
  • Built custom WordPress plugins, including multi-site authentication and Mailchimp integrations.
Vue.jsES6WordPressSCSS

Before that, from 2011 to 2016, I built websites for a range of clients and taught myself the full stack along the way.

Backend & platform

Multi-tenant systems that have to stay cheap, fast and correct while the number of tenants keeps climbing.

Python & Django

Multi-tenant SaaS backends on Django and DRF, with Graphene for GraphQL.

Django 5.2DRFGraphene

Data & caching

Query and write tuning, replication, time-series analytics, connection pooling.

PostgreSQLTimescaleDBRedisPgBouncer

Caching & performance

Response caching with stale-while-revalidate, stampede locks and entity-scoped invalidation.

RedisSWRCDN

APIs & async work

GraphQL and REST, WebSockets over Redis channels, background jobs on Celery.

GraphQLRESTCeleryWebSocket
MeNEW platform backend

MeNEW platform backend

The Django core behind ~700 venues: GraphQL API, background translation, subscriptions and billing.

−80%
DB writes at peak
−50%
queries at peak
8 → 2
backend nodes
  • Designed a Redis response cache with stale-while-revalidate, lock-based stampede protection and entity-scoped invalidation.
  • Found and fixed a cache-key poisoning bug where responses were keyed by operation name alone.
  • Cut writes by touching visit timestamps only on real change and bundling updates into single atomic transactions.
  • Split guest analytics into a dedicated TimescaleDB database with campaign attribution and session lifecycle handling.
DjangoGraphenePostgreSQLTimescaleDBRedisCelery
Subscriptions & billing engine

Subscriptions & billing engine

A rebuild of how plans are sold, renewed and provisioned, after paid invoices were failing to deliver.

  • Redesigned the inventory and possession model behind every plan and add-on.
  • Wrote the dating rules for renewal, upgrade and downgrade, with credit-rollover quoting.
  • Separated invoice bookkeeping from provisioning so a bookkeeping error can no longer swallow a delivery.
  • Added superadmin deal-date adjustment for support cases.
DjangoPostgreSQLCelery

AI translation pipeline

Menus translated into six or more languages automatically, with per-language progress and recovery.

3×
faster end to end
6+
languages, RTL included
  • Migrated from the OpenAI Assistants API to the Responses API.
  • Added per-language progress tracking, logging and a stuck-pass recovery and resume flow.
  • Replaced a legacy feature ladder with a language-quota entitlement model.
OpenAI Responses APICeleryDjango signals

AI immersive platform backend

The Django DRF service powering OZure's avatars, voice and retrieval.

  • Set up LangChain with PDF loading, splitting and embeddings, connected to Qdrant for vector search.
  • Integrated ElevenLabs TTS and Deepgram STT for the audio layer.
  • Built customizable prompt templates so one avatar could take on different professional personas.
DjangoOpenAILangChainQdrant
TAB customer dashboard

TAB customer dashboard

A business management system covering most day-to-day operations for TAB and MEP companies.

  • Sales, accounting, test sheets, reporting and financial management in one platform.
  • Automated report generation from user-supplied documents and data.
  • Integrated S3-compatible storage for file handling.
  • Built an estimation tool that reads mechanical drawings and drafts work-cost records.
DjangoDRFPDF generationWasabi S3
MLM profit engine

MLM profit engine

A multi-level profit distribution system built to absorb high daily calculation volume.

  • Scalable distribution across unlimited levels using Celery and async logic.
  • Incoming and outgoing email converted into support tickets automatically.
  • Built Earner, which monitors on-site engagement and rewards activity with profit tracking.
DjangoPostgreSQLCelery

Infrastructure & DevOps

I own the bill as well as the uptime, so most of this work is about getting more reliability out of less hardware.

Containers & delivery

Reworked Dockerfiles for three services with internal mirrors, pinned Poetry and non-root users.

DockerComposeGitLab CI

Linux & runtime

NGINX, uvicorn worker recycling, Celery memory leaks, cron automation, cPanel and WHM.

LinuxNGINXuvicorn

Resilience

Active and passive servers with PostgreSQL replication, plus a fully independent backup stack.

ReplicationFailoverCSFSSL

Observability

Sentry across backend and both frontends with tuned sampling, plus a durable Telegram alert outbox.

SentryGrafanaPrometheus

How the infrastructure got here in MeNEW

  1. 2026

    Emergency static fallback

    HTML and CSS of every active venue published to an independent domain, CDN and server, reachable by QR code during a total outage.

  2. 2025

    Cost and redundancy rebuild

    Monthly spend cut from ~$1,500 to ~$800 while the platform grew 40%, with a passive replica added rather than pocketing the saving.

  3. 2025

    PgBouncer outage resolved

    Connection exhaustion fixed with concurrency limits and refusal caching; Celery client memory leaks closed.

  4. 2024

    Database optimization

    PostgreSQL tuned for high-volume queries and replication set up for high availability.

  5. 2023

    Containerization

    Docker Compose workflows standardized across development and production.

  6. 2018

    Cloud server management

    Provisioning and backups on DigitalOcean and AWS.

  7. 2015

    cPanel / WHM administration

    CentOS optimization and CSF firewall configuration.

Frontend

Interfaces that stay responsive on a mid-range phone, in all languages and in both reading directions.

React & Next.js

App Router, server rendering, internationalized routing and runtime theming.

React 19Next.js 15TypeScriptnext-intl

Design systems

One system across panel pages, with container-query responsive layouts.

Tailwind v4SCSSStorybook

Rendering performance

Memoization, virtualization, progressive blur and off-screen video pausing for mobile.

Virtualizationdnd-kitblurhash

Vanilla JavaScript

Framework-free, class-based components and a custom HTTP abstraction layer.

ES2025Modular JS
MeNEW admin panel & guest app

MeNEW admin panel & guest app

A full redesign of the operator panel and a migration of the customer-facing menu to the App Router.

20 → 120 fps
menu editor
20+
pages rebuilt
6+
languages, RTL
  • Rebuilt settings, dashboard, finance, staff and onboarding pages on a single design system.
  • Moved the guest app from Pages Router to App Router with next-intl, SSR, slug routing and dynamic sitemaps.
  • Replaced the legacy editor with a kanban-style drag-and-drop Menu Builder on dnd-kit.
  • Built a gallery lightbox, styled short-link QR sharing and occurrence-aware routing for repeated items.
Next.jsReactTypeScriptTailwind v4dnd-kitPWA
VirtualMe avatar platform

VirtualMe avatar platform

A Next.js and React Three Fiber platform for talking 3D avatars driven by real speech.

  • Adapts frame rate to the viewer's graphics capability.
  • Loads an entire project — title, avatar voice, LLM instance — from the URL alone.
  • Maps Azure text-to-speech blend data onto GLB shape keys for human-like visemes.
  • Speaks and understands multiple languages.
Next.jsReact Three FiberWebGLAzure TTS

Immersive & AI

3D on the web, generated from photographs, and the GPU pipelines that make it fast enough to be useful.

3D & scene work

Mesh optimization, texture baking, draw-call reduction and compression.

Three.jsWebGLBlenderGLB

Gaussian Splatting

Single images reconstructed into explorable environments, exported for the web.

SplattingSAM-3DPyTorch

AI integration

LLM-driven avatars, speech in and out, and image generation pipelines.

OpenAIElevenLabsAzureComfyUI

BeeStoon AI

Turns a single photograph into a 3D environment you can walk through in a browser.

5 min → 30s
scene generation
−50%
exported asset size
1
photo needed, not 12
  • Gaussian Splatting and mesh reconstruction behind a plain web viewer.
  • Pipelines built around current research repositories including SAM-3D-Objects.
  • Texture-baked GLB exports produced automatically instead of by hand.
  • GPU infrastructure on Linux with CUDA, PyTorch, Mamba and Docker.
Gaussian SplattingPyTorchCUDAGLB

Storyboard Studio

Cinematic pre-production on the web: compose scenes, sequence shots, export breakdowns.

  • Drag-and-drop scene composition with sequencing and animatic workflows.
  • Asset libraries, shot configuration and production breakdowns.
  • Storyboard cards generated with open-source vision and image models on ComfyUI.
Next.jsComfyUIVLM

OZure immersive platform

Multilingual voice commands, LLM-powered avatars and real-time scene teleportation.

<10 → 50–60 fps
immersive scenes
  • Real-time voice interaction built on Whisper, ElevenLabs and Azure Face API.
  • Scene teleportation and avatar animation in PlayCanvas.
  • Frame rate lifted from under 10 to a steady 50–60 through draw-call and asset work.
PlayCanvasThree.jsWebXROpenAI

BeeStoon and Jodo run on on-demand GPU servers, so they may be idle when you visit.

Looking for my next role

Backend, platform or infrastructure work on a product that has real users and real constraints. Remote anywhere, and I can relocate.

Download my CV

Three pages: the full history, with the numbers above in context.

Download CV

Send a message