Mohammed Faisal Parvez — ML & Edge AI Engineer / Independent Researcher, Hyderabad, India.

I study how vision-language models fail — and make them run where they shouldn't. My work sits at the intersection of VLM reliability & calibration, extreme quantization (1-bit), edge inference, and sim-to-real transfer: benchmarking frontier vision-language pipelines on cloud A100s, then compressing what survives down to CPU/ARM edge hardware.

The through-line: failure is data. Refusal behavior, silent training failures, and sim-to-real gaps are measurable, family-specific properties — and picking models by failure mode beats picking them by leaderboard rank.

Download Resume (PDF)↓