Home/Expertise
Real-time systems
WebSocket backends for live code translation and IoT race control, with published latency and concurrency numbers. Written from shipped work by Muhammad Huzaifa Shahbaz, AI Systems Engineer at Digital Dividend Global.
What I build
- Live translation and execution platforms where the user sees output as the program runs.
- IoT control planes that broadcast race state, staging sequences, and safety flags to connected devices.
Architecture
- PseudoCompiler uses FastAPI and WebSockets on AWS. Lexical translation from AQA Pseudo to Python 3 streams results to the browser, with an AI chat path for errors.
- Isolated Docker runtimes execute student code. MongoDB stores sessions. LangChain coordinates the assistance steps.
- Staging Pro uses Django REST Framework, Django Channels, PostgreSQL, and Redis. Coordinators manage concurrent events; devices receive telemetry and flags.
- Race progression is logged as immutable events so the live broadcast and the audit trail are not the same write path.
Production constraints
- PseudoCompiler is published at 50,000+ concurrent users and under 50ms translation latency, with AWS cost reduced by about 35%.
- Staging Pro processes live telemetry under 100ms so a safety flag is not a batch job.
- Fan-out is the hard part: one coordinator action has to reach every connected device without turning the database into the message bus. Redis sits in that path.
Stack
FastAPI · Django · Django Channels · WebSockets · Redis · PostgreSQL · AWS · Docker
Tradeoffs
- WebSockets give the latency these products need and cost more to operate than request-response HTTP, especially around reconnects and sticky sessions.
- Running untrusted student code means the isolation boundary (containers) matters as much as the translation accuracy.
Projects
- PseudoCompiler — Live AQA Pseudo to Python translation and execution.
- Staging Pro 2.0 — IoT race control, telemetry, and safety-flag broadcast.
External verification
Questions
What concurrency and latency numbers are published?
PseudoCompiler: 50,000+ concurrent users and under 50ms translation latency. Staging Pro: live telemetry under 100ms.
Which stacks run those systems?
PseudoCompiler is FastAPI, WebSockets, and AWS. Staging Pro is Django, Django Channels, PostgreSQL, and Redis.