Samuel Hussain

BA (Hons) Computer Science.

Relevant modules: Algorithms I & II, Operating Systems (UNIX), Object-Oriented Programming, Databases, Software & Security Engineering, Machine Learning.

Activities: Chess, Competitive Programming, Cambridge University Islamic Society, Churchill Football Club.

A Levels: A*A*A*A* in Mathematics, Further Mathematics, Physics and Chemistry.

Awards: UKMT Senior Maths Challenge Gold and Senior Kangaroo Merit (both #1 in college), BPhO Computational Challenge.

• Placed 2nd of 38 invited teams in a week-long competition on algorithmic game theory, winning a $2,500 prize.
• Proposed and built a training environment that reconstructed opponents' algorithms from match logs, allowing the team to stress-test against realistic opposition rather than guessing.

• Selected as 1 of 50 from around 1,000 applicants. Built a voice-first crisis reporting application in a day and presented it to senior Google engineers.
• Streamed microphone audio from a React AudioWorklet over a WebSocket to a FastAPI backend owning the Gemini Live session, keeping conversational state server-side.
• Added interruption handling via server-side voice activity detection, and made incident extraction idempotent so retries never created duplicates in the responder queue.

• Built an exchange-style matching engine with price-time priority, supporting Good-Till-Cancel, Fill-and-Kill and Good-For-Day order types.
• Designed thread-safe order expiry: a background thread cancels Good-For-Day orders at market close, coordinated through a mutex, condition variable and atomic shutdown handshake, with RAII locking throughout.
• Microbenchmarked per-operation latency on a 100,000-order book, identifying and removing a pathological per-add allocation.

• Built a leakage-safe win-probability model for Counter-Strike 2 matches with XGBoost and scikit-learn, ensuring every feature used only data available before the match predicted.
• Benchmarked against the production Elo rating with paired-bootstrap confidence intervals, DeLong tests and a walk-forward backtest over 5,000 out-of-sample matches, finding the baseline near-optimal.
• Shipped it as a live Django application with a caching layer making repeat predictions around 100 times faster, fed by a checkpointed crawler with deduplication and rate-limit backoff.

• Administered Windows Server environments supporting secure cross-branch data exchange, and configured a Squid web proxy on Ubuntu Server to reduce external bandwidth usage.
• Improved retrieval times for cached content by 60% by refining server-side caching and security rules.