Aryan Lokesh
Graduate Software and AI Engineer · United Kingdom
Profile
MComp Computer Science with Artificial Intelligence graduate with project experience in Android development, Rails applications, audio machine learning, GPU programming and numerical computing. UK Graduate visa holder.
Education
University of Sheffield · 2022-2026MComp Computer Science with Artificial Intelligence
Four-year integrated master's, 2:1
Final-year average: 70.1%. Selected results: Parallel Computing with GPUs 79; Darwin Project 75; Speech Technology 74.
Cambridge International A Levels (June 2021): Information Technology A*, Mathematics A, Chemistry B, Physics B; English Language B (AS Level).
Technical skills
Languages: Python, C/C++, Kotlin, Ruby, SQL
Frameworks and tools: PyTorch, Ruby on Rails, PostgreSQL, CUDA/OpenMP, Jetpack Compose, Room, NumPy/CuPy, SciPy, Git, Linux
Selected university projects
Ruby on Rails 8, PostgreSQL | Genesys team project
- Year 4 team project: a productivity platform with personal/shared tasks, team projects, Pomodoro, notifications, gamification and role-based analytics.
- The team codebase includes RSpec/Capybara tests, performance examples and CI configuration for testing, security and linting.
Kotlin, Jetpack Compose, Room, Google Maps | Six-person team
- Implemented property creation, viewing, editing and deletion, maps/geocoding, proximity sorting, supplier assignment, internal image handling and related tests.
- Connected Compose screens to Room persistence and map/list navigation, with alphabetical sorting when live location was unavailable.
Python, PyTorch, Google HeAR, HPC | HeAR workstream within team research
- Developed patient-level multiple-instance learning over frozen HeAR cough embeddings, with grouped five-fold cross-validation, sensitivity-targeted thresholds and bootstrap analysis.
- The DS2 symptom-assisted configuration reached 0.830 AUROC in retrospective internal cross-validation; this was screening research, not clinical validation.
C/C++, CUDA, OpenMP, Nsight | Individual coursework
- Implemented and optimised Count Gliders, Histogram and Emboss in OpenMP and CUDA against supplied serial CPU references.
- Checked correctness and profiled execution with Nsight; compared end-to-end timings including allocation, host-device transfers, kernel execution and cleanup.
Python, NumPy/CuPy, PyTorch, SciPy | Individual dissertation
- Implemented CPU COS pricing, batched CuPy GPU pricing and bounded L-BFGS-B calibration of the Heston stochastic-volatility model to filtered SPX option contracts.
- Trained a ten-feature PyTorch surrogate on 600,000 synthetic samples and compared calibration runtime and pricing error across CPU, GPU and surrogate paths.
Python, PyTorch, scikit-learn | Individual coursework
- Compared MLP, recurrent and temporal-convolutional voice-activity detectors; the selected TCN achieved 7.73% EER over 495,486 cross-corpus test frames.
- Separately implemented Gaussian mixture models for vowel classification and closed-set speaker identification, comparing mixture complexity and utterance-level decisions.