ML Engineer · LLM Evaluation & ML Systems · Bristol
I measure what ML systems actually do, and whether the numbers hold up. MSc in High Performance Computing with Data Science from EPCC, University of Edinburgh, now based in Bristol.
BTech in AI, then an internship at ISRO's National Remote Sensing Centre where I built AetherVision, a ConvLSTM lightning-prediction model trained on weather-model output and 163,000+ recorded strikes, submitted as a research prototype to ISRO's internal research library. I turned down a job offer to do the MSc because I wanted to understand systems at a deeper level.
MSc at EPCC, Edinburgh: parallel programming on ARCHER2, GPU computing, and a dissertation on catching container memory leaks early, with eBPF collectors, TimescaleDB and ten detectors compared at one fixed false-alarm budget. Supervised by Ben Carpenter.
Outside the MSc: KVScope, two technical reports measuring KV-cache memory on H100 and H200; a streaming jailbreak detector that I audited and corrected myself; an LLM serving stack I'm load-testing; and merged OpenTelemetry contributions to Jaeger, the CNCF tracing project.