open to opportunities

Mohamed Hosny

Senior Software Engineer / Software Architect

Cairo, Egypt · UTC+2 · remote

I turn messy enterprise systems into reliable products fast. From national infrastructure and regulated fintech platforms to AI-native builds, I design systems that survive real production pressure.

  • 8+ yrsin production
  • 6industries
  • 40×calibration throughput
  • IEEEpublished research
Clientweb / mobileAPIREST + WSServices.NET / NodeAI Agenttools + RAGPostgresLLMone contractidempotent
the shape most of my work takes

Capabilities

What I Build With

Backend

Services built for regulated domains, real traffic, and on-call support.

C# / .NET CoreNode.jsPythonFastAPIMicroservicesOAuth2

Frontend

Interfaces that stay maintainable after the third team touches them.

TypeScriptAngularReactNext.jsVue

AI

Agentic systems, orchestration, and LLM architecture that survives a budget review.

Claude APIOpenAIvLLMRAGMulti-AgentTool Use

Cloud / DevOps

Infrastructure and delivery flows that keep teams shipping under pressure.

AzureAWSDockerGitHub ActionsTerraformRedis

Case studies

Things I built, and what happened

5 of 11. Each one links to the full write-up with the decisions, trade-offs, and a working demo.

Betting Wallet Redesign

A wallet service where two bets placed at the same instant could both spend the same balance.

Row lockcheck and debit in one transaction

Problem

Under load, concurrent bets on one account raced each other.

What I built

Moved the check and the debit inside a single transaction that takes a lock on the wallet row before reading it.

What it cost

Writes to a single wallet now serialize, so a hot account is a throughput ceiling no number of instances can raise.

  1. Bet Request
  2. Wallet Lock
  3. Check & Debit
  4. Commit
  5. Settlement

Betting Wallet Redesign — request path

C#/.NETSQL ServerTransactionsConcurrencyMicroservices
Read the case study →

Computer-Use Agent

Agents that read the screen as pixels and drive legacy desktop software that offers no API to call.

< 2sscreenshot → action

Problem

Businesses need to automate complex desktop workflows — filling forms, navigating legacy software, extracting data from screens — but traditional RPA is brittle and breaks when UIs change.

What I built

Built a production platform on Claude Computer Use API with a Python/FastAPI backend.

What it cost

Vision-driven control is slower and costlier per step than scripted RPA, and it is non-deterministic — every action needs a verification read-back before the next one is safe.

  1. Screen Capture
  2. Vision Analysis
  3. Action Planning
  4. Execution Engine
  5. State Recovery

Computer-Use Agent — request path

Claude APIComputer UsePythonFastAPIVision AITool Use
Read the case study →

Enterprise Payment Microservices2023–now

Payment services split along the boundaries that matter — gateway, payments, ledger, notifications — carrying millions of records a day.

MillionsRecords / Day

Problem

High-volume payment traffic needed dependable services and fast production fixes — millions of records a day flowing through gateway, payments, ledger, and notification paths, where downtime or backlog has direct financial impact.

What I built

Delivered .NET Core microservices with clear service boundaries, Redis-backed hot paths, and Docker-based deployment, plus the hotfixes and operational support that keep a high-throughput payments platform healthy under real load.

What it cost

Clear service boundaries cost cross-service consistency — transactions stay inside a single service and the ledger reconciles afterwards.

  1. API Gateway
  2. Payments Service
  3. Ledger Service
  4. Notifications

Enterprise Payment Microservices — request path

.NET CoreMicroservicesDockerRedisSQL Server
Read the case study →

Smart Meter Calibration System2018

A multithreaded calibration pipeline that took smart-meter manufacturing from one meter per cycle to forty.

40×1 → 40 meters per cycle

Problem

Manufacturing throughput was capped by a slow, sequential calibration workflow — each meter processed one after another, leaving expensive production lines waiting.

What I built

Re-architected calibration around a multithreaded, batch-oriented pipeline tuned to the production hardware.

What it cost

Parallelising the line meant giving up a simple failure model: one meter faulting mid-grid must not take the other thirty-nine with it, so per-slot error isolation and recovery became the real work.

  1. Batch Intake
  2. Parallel Calibration
  3. Result Streaming
  4. Verification

Smart Meter Calibration System — request path

C#/.NETMultithreadingSQL ServerManufacturing
Read the case study →

Healthcare EDI & HL7 Integrations2022

Dynamic 835/837 EDI and HL7 parsers that turn messy, multi-vendor healthcare interchange into reliable structured data.

835/837EDI Formats Parsed

Problem

Clinical and billing systems exchange data as dense X12 EDI (835 remittance, 837 claims) and HL7 messages that vary by vendor and break rigid parsers.

What I built

Built schema-driven parsers that tokenize the X12 stream, resolve segment definitions dynamically, validate loops and control numbers, and normalize everything into internal claim and remittance models — then route the results through Iguana, a fax gateway, or internal services.

What it cost

A schema-driven parser is slower and harder to debug than a hand-written one per vendor, and it adds a schema registry to maintain.

  1. Segment Tokenization
  2. Schema Resolution
  3. Validation
  4. Transformation
  5. Routing

Healthcare EDI & HL7 Integrations — request path

C#/.NETHL7EDI 835/837X12IguanaSQL Server
Read the case study →

See all 11 case studies →

System design

Three shapes I draw a lot

Whiteboard defaults I reach for before reaching for anything clever.

The boring one that pays the bills

Most business software is this shape. The work is in the boundaries, not the boxes.

  1. Userbrowser
  2. APIREST
  3. Backendservices
  4. DatabasePostgres
Validate at the edge, keep transactions inside one service, and the rest stays simple.

Retrieval before generation

An agent is only as good as the context it is handed. Retrieval is the product; the model is a dependency.

  1. Appchat / task
  2. AI Agenttool use
  3. RAGvector store
  4. LLMvLLM / API
Cache aggressively and cap the loop — cost lives in the number of round trips, not the prompt.

Real-time without the pain

Sockets are easy until you run two instances. Shared state is what makes it survive a deploy.

  1. ClientWebSocket
  2. Gatewaysticky-free
  3. Backendhandlers
  4. Redispub/sub
Fan out through Redis so any node can serve any client, and a restart costs nothing.

Experience

Where the scar tissue came from

Government infrastructure, Saudi fintech regulation, healthcare interoperability, payments, manufacturing, and AI — in that order.

  1. Senior Software Developer

    Enterprise payments platform · remoteMar 2023 – Present
    • Microservices enterprise payment solution — .NET Core, Azure, SQL Server
    • Optimized large-scale data processing handling millions of payment records
  2. Senior Full-Stack Developer

    Healthcare & payments consultancy · remoteJan 2022 – Mar 2023
    • Built modular ISO 8583 solutions with dynamic 835/837 EDI healthcare parsing
    • Integrated Fax, HL7 via Iguana, and internal healthcare systems
  3. Full-Stack Session Lead

    Engineering education · remoteApr 2021 – Aug 2021 · alongside the day job
    • Mentored 50+ full-stack nanodegree students in Node.js, Express, React
    • Weekly meetups, code reviews, and one-on-one sessions guiding certification
  4. Senior Software Developer

    Healthcare & fintech consultancy · remoteApr 2020 – Jan 2022
    • Full-stack healthcare & fintech platforms — .NET Core, Node.js, Angular, AWS
    • Secure OAuth2.0 authentication with IdentityServer for multi-tenant apps
  5. Senior Software Developer

    Smart-meter manufacturing · on-siteMay 2017 – Apr 2020
    • Calibration system that took the line from one meter per cycle to forty
    • Low-level DLMS & IEC 1107 protocols for smart meter manufacturing
  6. Software Developer

    Enterprise GIS · on-siteOct 2016 – May 2017
    • WPF/MVVM applications with DevExpress for enterprise GIS-integrated solutions
    • Legacy system modernization and improved GIS data processing responsiveness

Contact

Need someone who can enter complexity fast and ship?

If you are untangling a hard platform, replacing brittle workflows, or building an AI-native system from scratch, let’s talk.