Prashant Vithule

Software Engineer

Tiroda, Maharashtra, India1 yr 9 mos experience

Key Highlights

Expert in AI systems and LLM optimization.
Proven track record in high-performance computing.
Skilled in advanced vectorization techniques.

Stackforce AI infers this person is a specialist in AI and high-performance computing.

Contact

Skills

Core Skills

Large Language Models (llm)NumaSve

Other Skills

Advanced Vector Extensions (AVX)SIMDOpenMPThread affinity

About

Passionate about AI systems, LLM & LMM inference optimization, and high-performance computing. Worked on quantized kernel optimization, speculative decoding, threading affinity, NUMA-aware design, and SIMD/SVE acceleration for large language models. Experienced in C, C++, CUDA, OpenMP, LLVM, NUMA, and GPU programming.

Experience

1 yr 9 mos

Total Experience

1 yr 8 mos

Average Tenure

1 mo

Current Experience

Amd

Sr. Software System Designer

Apr 2026 – Present · 1 mo · Bengaluru · Hybrid

SVEAdvanced Vector Extensions (AVX)SIMDOpenMPLarge Language Models (LLM)Thread affinity+1

Fujitsu research

Software Engineer

Jul 2024 – Mar 2026 · 1 yr 8 mos · Bengaluru · Hybrid

1. Implemented SVE-based vectorized dot product kernels for quantized formats (Q2_K, Q3_K, Q4_0, Q8_0) in llama.cpp, achieving up to 3× throughput improvement on 512-bit architectures.
2. Implemented a lookahead speculative decoding framework in llama.cpp, enabling faster inference.
3. Designed NUMA-aware execution and optimized thread affinity to improve CPU utilization and reduce cross-node memory latency.

SVEAdvanced Vector Extensions (AVX)Large Language Models (LLM)