Dr. Aris Thorne · AI / ML
HuggingFace Weights
PyTorch · FlashAttention-2 · CUDA

Training Efficient Neural Networks & Scaling Large Language Model Inference

Principal AI Systems Researcher & LLM Optimization Engineer

Specializing in LoRA/QLoRA parameter-efficient fine-tuning, KV-cache quantization, and custom Triton kernels. Bridging cutting-edge deep learning theory with sub-10ms production inference systems.

Inspect Model Weights Research Citations

AI Production Benchmarks

Validated inference speedups and model deployment metrics

4.2x
Inference Speedup via Triton
1.4M+
HuggingFace Model Downloads
70B
Largest LLM Quantized & Deployed
98.4%
Evaluation Accuracy Benchmark

Open Source Weights & Quantized Engines

Replicable deep learning models with reproducible benchmark scripts

OmniQuant-70B Engine

4-bit AWQ quantized Llama-3 variant optimized for high-concurrency consumer GPU serving with FlashAttention-2 integration.

PyTorchvLLMAWQCUDAPython
Live Demo Code

Triton-SparseKernel

Custom OpenAI Triton kernel executing block-sparse matrix multiplication with 38% latency reduction over cuBLAS.

TritonGPU OptimizationDeep Learning
Live Demo Code

AgenticRag Orchestrator

Self-correcting retrieval augmented generation system utilizing hybrid dense/sparse embeddings and rerankers.

LangChainQdrantFastAPIDocker
Live Demo Code

Initiate AI Research Collaboration

Available for model optimization contracts, specialized LLM fine-tuning, and research advisory.

aris.thorne@craftfolio-ai.org
San Francisco / Remote
Open for selective AI research & consulting contracts