
Founder
Inferno · Jan 2026 – Present
- • Built InfernoRT, an AI inference runtime that outperforms llama.cpp across every benchmark configuration where both runtimes successfully complete execution.
- • Enabled workloads that fail on other runtimes due to memory constraints, including 8B-class models within a 4 GB RAM budget and 70B-class models within 16 GB RAM.
- • Delivered higher decode throughput and lower time-to-first-token than other runtimes while preserving output correctness under our validated benchmark suite.
- • Achieved almost 2x the imporvements in TTFT than llama.cpp
- • Built a production benchmarking framework spanning 480 workloads to validate latency, throughput, correctness, and memory behavior against the industry baseline.
- • Filed core IP for deterministic memory management enabling efficient local AI inference.










