usama@portfolio — zsh

0%
11+ Years EngineeringArchitect & Lead Engineer · Televox AIMicrosoft Certified · Azure Solutions Architect Expert

ArchitectingScalablePlatforms&AgenticAISystems.

Software architect with 11+ years designing, building and operating full-stack platforms across agentic AI, real-estate CRM, fintech and analytics — from a sub-second voice AI pipeline in production to a self-hosted LLM inference server. Works AI-first, with Claude Code in daily use.

View ProjectsDownload CV
0.8–1.0sP50 Voice Response
5Services Run Solo
70–84Local LLM Tokens/s
~40%CRM Revenue Lift
About the Architect

Engineering platforms with an architect's lens.

I'm Usama — a software architect with 11+ years designing, building and operating full-stack platforms across agentic AI, real-estate CRM, fintech and analytics.

I'm the architect and lead engineer of Televox AI, a multi-tenant voice and agentic AI SaaS running in production with live paying customers at sub-second p50 response — designed, shipped and operated end to end, including customer discovery and support. Before that I led back-end and front-end architecture and mentored engineering teams through seven years at Optimageeks, and delivered front-end work on a CFD and forex trading platform serving millions of traders.

I work AI-first: Claude Code daily across design, implementation and testing, with every piece of model-generated code reviewed for correctness, performance, tenant safety and idempotency before it ships. I also built and run my own LLM inference server on Apple silicon as private AI coding infrastructure.

Microsoft Certified: Azure Solutions Architect Expert, Azure AI Apps and Agents Developer Associate, and Azure Administrator Associate. Delivery experience across the US, Australia, UAE and Pakistan. UAE resident — no sponsorship required.

UAE · PakistanOpen to remote workUAE resident · no sponsorship required
Microsoft Certified · 3 Azure credentials
Engineering Arsenal

Architecture

Multi-tenant SaaS design and tenant isolation, microservices, real-time and event-driven systems, REST API and schema design, observability and instrumentation.

AI Engineering

LLM integration (OpenAI, Anthropic, Gemini, Azure OpenAI) with tool calling and agent orchestration, RAG with pgvector, Azure AI Foundry, evaluation and guardrails, self-hosted inference on MLX.

Real-Time & Full-Stack

WebRTC (LiveKit), WebSockets, SSE, Deepgram speech and Twilio SIP telephony — built in TypeScript and Python on Next.js, React, NestJS and FastAPI.

Leadership & Delivery

Technical direction and architecture ownership, code review and mentoring, customer discovery, cross-functional delivery, Agile/Scrum and incident triage.

The Journey

Experience Timeline

From engineering teams in Lahore to a production agentic AI SaaS — every step engineered forward.

Late 2025 — Present

Architect & Lead Engineer

Televox AI · televox.ai · live paying customers

Sole engineer on a multi-tenant agentic AI SaaS, architected and operated end to end: a single agent configuration answers inbound PSTN calls over a Twilio Elastic SIP trunk, web voice over LiveKit WebRTC, and an embeddable website chat widget.

  • Designed the real-time pipeline (Silero VAD → Deepgram Nova-2 STT → OpenAI with tool calling → Deepgram Aura-2 TTS) with barge-in and semantic turn detection — p50 response held at 0.8–1.0s under load
  • Own the production estate: five services, a stateless voice-worker fleet on Fly.io with LiveKit load balancing, admission control, and per-turn STT/LLM/TTS latency and cost instrumentation
  • Per-agent knowledge bases with pgvector RAG, HMAC-signed webhooks, real-time Cal.com booking, SMS actions, warm call transfers, and outbound campaigns with TCPA safeguards
  • Run customer discovery and support directly; self-serve telephony provisioning, plan enforcement and billing
May 2026 — Present

Principal App Developer

Medical Guardian · US (Remote)

Principal developer on React Native applications and the API integration layer for a US medical alert and connected-care provider.

  • Define front-end structure and API contracts for new features with product, back-end and QA
  • Work within the reliability and compliance expectations of a healthcare product where device and response workflows cannot fail
Apr 2024 — May 2026

Senior Software Engineer

Pepperstone · Australia (Remote)

Front-end engineering on a global CFD and forex trading platform serving millions of traders, partnering with a C# back-end team to deliver trading features and conversion flows.

  • Built React and TypeScript interfaces for the crypto trading experience, including real-time market data delivered over WebSockets
  • Implemented secure authentication meeting financial-industry compliance requirements
Sep 2022 — Mar 2024

Senior Software Developer

Modulus · US (Remote)

Senior developer across the product portfolio, including Anzer — turning business requirements into system design and specifications.

  • Designed and ran integration and performance testing across multiple environments to protect system stability
  • Coordinated priorities across cross-functional teams and led triage of development and production issues
Jan 2015 — Aug 2022

Software Engineer → Senior Software Engineer → JavaScript Full-Stack Developer

Optimageeks · Lahore, Pakistan

Seven years leading back-end and front-end architecture for core web applications — database schema design, server-side API design and server provisioning — across client platforms including Optima CRM, My Lobby, LodgeIntel and 100Bet.

  • Led front-end architecture and designed the user-facing interactions and flows across client platforms
  • Mentored junior engineers through code review and pairing, and set the review standards the team worked to
  • Designed and built REST APIs for system integration and inter-service data exchange, integrating internal applications with third-party systems
2008 — 2015

BSc Electronics and Communication

University of Lahore

Bachelor of Science in Electronics and Communication, University of Lahore (Sep 2008 – Mar 2015) — the foundation for a decade-plus of software engineering and systems thinking.

  • Transitioned from electronics and communication systems into full-stack software engineering
Active Development Lab

Currently Building

Active Development

Televox AI

Multi-tenant voice and agentic AI SaaS with live paying customers — one agent config answering PSTN calls, web voice and website chat, architected and operated end to end as sole engineer.

Latest: Admission control and per-turn cost instrumentation across the voice-worker fleet.

In Daily Use

Local LLM Inference Server

Headless MLX server on Apple silicon (M4 Pro, 24 GB) serving open-weight models through OpenAI- and Anthropic-compatible APIs over a private Tailscale network — launchd bring-up, persisted GPU wired-memory limit, crash watchdog.

Latest: Resolved a GPU out-of-memory failure with a 4x smaller KV-cache at 64K context; seven days unattended uptime.

Active Development

Medical Guardian Apps

Principal developer on the React Native apps and API integration layer for a US medical alert and connected-care provider.

Latest: Defining front-end structure and API contracts for new features with product, back-end and QA.

Tech Stack

Skills & Engineering Stack

The technologies and disciplines behind every build — from pixel to pipeline.

Architecture

Multi-tenant SaaS & tenant isolationMicroservicesReal-time & event-driven systemsREST API designDatabase schema designScalable system designPerformance optimisationObservability & instrumentation

Languages & Frameworks

TypeScriptJavaScriptPythonPHPNode.js (Express, NestJS, Sails, LoopBack)ReactNext.jsReact NativeAngularFastAPILaravelYii2

AI Engineering

LLM integration (OpenAI, Anthropic, Gemini, Azure OpenAI)Tool calling & agent orchestrationPrompt engineering & structured outputRAG with pgvectorAzure AI FoundryAI evaluation & guardrailsCost & latency optimisation per conversationNLP & sentiment analysisSelf-hosted inference (MLX, quantised models, KV-cache)

Real-Time & Integrations

WebRTC (LiveKit)WebSocketsSSEDeepgram STT / TTSSilero VADTwilio SIP & telephonyOAuth2 / OIDCSigned webhooksThird-party API & CRM integrationn8n workflow automation

Data & Infrastructure

PostgreSQL (Supabase, pgvector)MySQLMongoDBRedisAWSAzureFly.ioDockerCI/CDNginx / ApacheSocket.ioGit / GitHubTailscale private networkingmacOS services (launchd, pmset, sysctl)Performance benchmarking

AI-Assisted Development

Claude Code daily — design, implementation, testingSpec-driven workflowsReviewing & validating model-generated codeCorrectness, performance, tenant safety & idempotency

Leadership & Delivery

Technical direction & architecture ownershipCode review & mentoringCustomer discovery & stakeholder engagementCross-functional deliveryAgile / ScrumIncident triage

Always shipping, always learning.

AI-first from design to test — Claude Code in the loop daily, with every generated line reviewed before it ships.

20+ technologies in production
Featured Engineering

Featured Projects

Flagship products and platforms — each engineered around a real problem.

Architect & Lead Engineer · Live SaaS

TELEVOX AI

Multi-Tenant Voice & Agentic AI SaaS

A multi-tenant voice and agentic AI SaaS with live paying customers, architected and operated end to end as sole engineer: one agent configuration answers phone calls, web voice and website chat — booking appointments, sending texts and capturing every lead.

Problem

SMBs miss 30–60% of inbound calls, and each missed call at a service business can be a $200–$5,000 lost job.

Solution

Designed a real-time pipeline (Silero VAD → Deepgram → OpenAI with tool calling → Deepgram TTS) holding 0.8–1.0s p50 under load, run by a stateless worker fleet on Fly.io that serves any agent from database config.

Phone, web voice & chat — one agent config
pgvector RAG, Cal.com booking, SMS & warm transfers
Outbound campaigns with TCPA safeguards
Per-turn latency & cost instrumentation
LiveKitTwilio SIPDeepgramOpenAISupabaseNext.jsPythonFly.io
Self-Hosted AI Infrastructure

LOCAL LLM INFERENCE SERVER

Private AI Coding Infrastructure on Apple Silicon

A headless inference server on Apple silicon (M4 Pro, 24 GB) serving open-weight models through OpenAI- and Anthropic-compatible APIs — published only on a private Tailscale network and in daily use for AI-assisted development.

Problem

Cloud LLM APIs meter every token and see every line of code. A private, always-on endpoint had to be fast enough to sit inside a coding loop and survive unattended on a single machine.

Solution

Designed, built and operate an MLX-based server behind an LM Studio daemon with launchd bring-up, a persisted GPU wired-memory limit and a crash watchdog — benchmarked at 70–84 tokens/s generation and ~450 tokens/s prompt processing.

70–84 tok/s generation · ~450 tok/s prompt
Seven days unattended uptime, daily use
GPU OOM fixed: 4x smaller KV-cache at 64K context
Private Tailscale-only network
MLXLM StudioTailscalelaunchdQuantised open-weight modelsOpenAI / Anthropic-compatible API
Conversational AI

CALLAGENT AI

Real-Time Conversational AI for Call Centres

A scalable, real-time conversational AI platform for call centres, customer support and telemarketing — automating inbound and outbound calls with AI agents, as lead developer.

Problem

Call centres rely on large human teams for repetitive inbound and outbound calls, limiting efficiency and consistency.

Solution

Built an AI agent system on NestJS, React and Docker integrating OpenAI, Gemini and real-time GPT services with Twilio call handling.

Conditional real-time call transfers
Cal.com appointment booking
Campaigns via Google Sheets & Zoho CRM
Admin panel: prompts, knowledge bases, webhooks
NestJSReactDockerTwilioOpenAIGemini
~40% Revenue Impact

OPTIMA CRM

Real-Estate CRM & Command Centre

A comprehensive cloud platform for real-estate agencies that replaced fragmented manual processes with one system — lead capture, centralised listings, digital offers, invoicing, commissions, rentals and a real-time BI layer — built on the MEAN stack.

Problem

Real-estate agencies lose leads and revenue to fragmented tools, manual listing management and inconsistent follow-up.

Solution

Engineered an all-in-one platform: lead capture from portals and social channels, centralised listings with automated portal syncing, and automated sales and rental operations.

Lead engine across portals & social channels
Listings hub with automated portal syncing
Digital offers, invoicing & commission tracking
Real-time BI dashboards & KPIs
MongoDBExpressAngularNode.js
FinTech · Millions of Users

PEPPERSTONE

Global CFD & Forex Trading Platform

Front-end engineering for a global CFD and forex trading platform serving millions of traders — React and TypeScript interfaces for the crypto trading experience, partnering with a C# back-end team.

Problem

A high-stakes trading platform needs fast, secure, compliant front-ends that millions of traders can rely on.

Solution

Built React and TypeScript trading interfaces with real-time market data over WebSockets, plus secure authentication and login systems meeting financial-industry compliance requirements.

Crypto trading front-end applications
Real-time market data via WebSockets
Secure authentication & login systems
Financial-industry compliance
ReactReact NativeTypeScriptC# BackendWebSockets
AI-Enhanced PropTech

MY LOBBY

Visitor Management & Facility Automation

A comprehensive visitor management system automating facility compliance and building security — replacing paper sign-in sheets with streamlined digital registration.

Problem

Buildings and offices track visitors on paper sign-in sheets, hurting security, compliance and productivity.

Solution

Built a MERN + React Native system with mobile prescreening, QR door access and an AI call agent for natural voice interactions.

Mobile prescreening & QR door access
Real-time sign-in/out dashboard
AI call agent & hot desking apps
Emergency notifications & queue management
MongoDBExpressReactNode.jsReact Native
Hospitality BI

LODGEINTEL

Hospitality Analytics & BI Suite

Cloud-based analytics and business intelligence for the hotel, resort and holiday park industries — optimising revenue, guest satisfaction and workflows.

Problem

Hospitality operators lack real-time visibility into revenue, occupancy and guest satisfaction across properties.

Solution

Engineered a MERN analytics suite with real-time sockets, MongoDB aggregation pipelines and interactive charting.

JWT authentication
Real-time data via Socket.io
MongoDB aggregation pipelines
Interactive graphs & charts
MongoDBExpressReactNode.jsSocket.io
EdTech

PLANETPOP

Interactive English Learning Platform

An interactive English learning platform built in Oxford, UK — combining structured lessons with rich media and video streaming.

Problem

Language learners need engaging, interactive content — static lesson pages fail to hold attention.

Solution

Delivered an Angular + Node.js platform with REST APIs and integrated video streaming for interactive lessons.

Interactive lesson experiences
Video streaming integration
REST API architecture
Built with Oxford, UK team
AngularNode.jsREST APIsVideo Streaming
E-Commerce

THE GOOD INSIDE

Headless Commerce for Touchstone Essentials

An e-commerce platform for Touchstone Essentials selling nutritional supplements — built on Next.js with headless WordPress and WooCommerce.

Problem

A growing supplements brand needed modern storefront performance without abandoning its WordPress content workflows.

Solution

Implemented a headless architecture: Next.js storefront over WordPress/WooCommerce with Elementor-driven content.

Headless WordPress architecture
WooCommerce checkout integration
Next.js storefront performance
Marketing-editable content
Next.jsWordPress (Headless)WooCommerceElementor
Frontend Engineering

100BET

iGaming Platform Redesign

Full redesign of an online betting site implemented from Figma specifications — reusable Angular Material components with third-party API integrations.

Problem

An established betting platform needed a complete visual overhaul without disrupting live operations.

Solution

Implemented the full redesign in Angular Material with a reusable component library and integrated third-party APIs.

Pixel-accurate Figma implementation
Reusable component library
Third-party API integrations
Full responsive redesign
AngularAngular MaterialFigmaREST APIs
Agency Web

DG-AGENCY

Digital Agency Portfolio Platform

Portfolio site for an IT and digital agency featuring the Perplex effect — built on Yii2 and Bootstrap.

Problem

A digital agency needed a distinctive portfolio presence that demonstrated its own front-end craft.

Solution

Built a Yii2 + Bootstrap site with the Perplex visual effect and a maintainable content structure.

Perplex visual effect
Custom Yii2 architecture
Responsive Bootstrap layouts
Agency case-study showcase
PHP (Yii2)BootstrapJavaScript
Selected Case Studies

Engineering Deep Dives

How key systems were architected, optimized, and shipped.

Televox AI Platform

Multi-tenant voice & agentic AI SaaS — sole engineer

The Problem

For most local businesses the phone is the #1 revenue channel and the most broken — 30–60% of inbound calls go unanswered, and each missed call at a service business can mean a $200–$5,000 lost job.

The Solution

Architected, built and operate the entire platform as sole engineer: a multi-tenant SaaS where one agent config answers PSTN calls, web voice and website chat — with per-agent knowledge bases, real actions (booking, SMS, warm transfers) and self-serve telephony, billing and plan enforcement. Customer discovery and support are run directly, so what small businesses describe becomes what ships.

Architecture
  • Real-time pipeline: Silero VAD → Deepgram Nova-2 STT → OpenAI with tool calling → Deepgram Aura-2 TTS, with barge-in and semantic turn detection
  • Stateless voice-worker fleet on Fly.io with admission control — any worker runs any agent from database config; LiveKit load-balances; number routing is a single indexed query
  • Supabase Postgres with row-level security on every table, pgvector RAG for large knowledge bases, and HMAC-signed webhooks
Performance
  • p50 voice response held at 0.8–1.0s measured under load; 8 concurrent calls with zero errors in stress tests
  • Prompt ordered for OpenAI caching — up to ~80% faster time-to-first-token when warm
  • Every turn instrumented end-to-end (STT, LLM TTFT, TTS TTFB) plus per-turn cost, surfaced through a latency-waterfall inspector

OUTCOME

Live in production at televox.ai with paying customers, five deployed services, measured unit economics and self-serve plans — architected, engineered and operated end to end by one engineer.

What I Do

Services Catalog

Focused engineering services that take products from idea to revenue.

Agentic AI & LLM Systems

Agent orchestration with tool calling, RAG on pgvector, structured output, evaluation and guardrails — on OpenAI, Anthropic, Gemini or Azure OpenAI, with cost and latency budgeted per conversation.

Real-Time Voice & Telephony

Voice agents over LiveKit WebRTC and Twilio SIP with Deepgram speech — barge-in, turn detection and sub-second response, instrumented per turn.

Multi-Tenant SaaS Architecture

Tenant isolation, event-driven services, schema and REST API design, observability — from data model to a stateless worker fleet on Fly.io, AWS or Azure.

Full-Stack Product Delivery

TypeScript and Python end to end — Next.js, React, React Native, NestJS and FastAPI on PostgreSQL, Supabase and Redis.

Self-Hosted & Private AI

Open-weight model inference on Apple silicon with MLX — quantised models, KV-cache and context budgeting, private networking and benchmarking.

Technical Leadership & Consulting

Architecture ownership and reviews, AI-assisted development workflows with Claude Code, mentoring, customer discovery and incident triage.

Ways to Work With Me

Engagement Models

Hiring for an architect or principal role, shipping a product, or looking for AI and architecture guidance — pick the path that fits.

Employment

Full-Time Role

Architect · Principal
For

Companies hiring architect- or principal-level engineering.

An architect-level engineer embedded in your team.

  • Software Architect / Principal Engineer positions
  • Full-stack ownership in TypeScript & Python
  • Agentic AI & real-time systems expertise
  • Mentoring & code review culture
  • UAE resident (no sponsorship) · remote worldwide
Most Flexible
Delivery

Contract / Project

Project-Based
For

Startups and companies with a defined product to ship.

End-to-end delivery — from architecture to launch.

  • Scoped milestones & timelines
  • Multi-tenant SaaS, CRM & analytics builds
  • Agentic AI: LLM tool calling, RAG, voice agents
  • Cloud deployment on Fly.io, AWS or Azure
  • Real-time & third-party integrations
  • Post-delivery support window
Advisory

Technical Consulting

Hourly · Retainer
For

Teams needing architecture guidance or AI strategy.

Senior judgment on demand, without a full-time hire.

  • Architecture & code reviews
  • AI / LLM strategy, evaluation & guardrails
  • Performance, latency & cost audits
  • AI-assisted development (Claude Code) enablement
  • Direct WhatsApp & Slack channels

* Open to full-time, contract and consulting engagements — remote worldwide, or on-site across the UAE. UAE resident, no sponsorship required.

Client Testimonials

Trusted by Founders & Teams

Signals from the people I have shipped with.

"The platform eliminated our manual listing errors entirely — and we saw a 40% increase in revenue after consolidating our operations onto it."
RA
Real-Estate Agency DirectorOptima CRM Client
"The AI agents handle our outbound campaigns end to end — real-time transfers, follow-up SMS, and booking straight into our calendar. It just works."
CC
Call Center Operations LeadCallAgent AI User
"Usama translated vague requirements into precise technical specifications and led triage across teams. Rare combination of architect and communicator."
PM
Product ManagerUS SaaS Company
"From mobile prescreening to QR access control, he owned the full stack — web, mobile, and the AI voice agent. Delivery was consistently ahead of schedule."
EM
Engineering ManagerFacilities Tech Platform
Initiate Contact

Let's build something exceptional.

Have a platform to scale, or want to put agentic AI to work in your product? Send a brief and I'll reply within 24 hours.

usamabinkhalil@gmail.com
+971 58 829 4334 (UAE)+92 336 404 2263 (PK)
UAE · Pakistan
Replies within 24 hours
Download CV (PDF)
Currently Available

Taking on a limited number of new SaaS & AI product builds this quarter — send a brief early to reserve a slot.