At DheyoAI, I built Eyoh’s scalable REST API gateway, supporting 140+ endpoints and 10K+ concurrent requests for multimodal AI workflows, collaborative storyboarding, LoRA fine-tuning, and payments.
I developed agentic workflows for prompt enhancement, script analysis, dataset captioning, and labeling, reducing manual editing effort. I also architected an automated LoRA fine-tuning pipeline with dataset management, queue processing, and SLURM scheduling.
I led the engineering team behind Nexor, a Rust static neural-network compiler and runtime with unified HIP and CUDA support. I implemented 30+ fused GPU kernels and validation benchmarks against PyTorch across AMD and NVIDIA hardware.
Previously at Kinara.ai, I configured LLVM for 50+ custom neural-network kernel instructions and designed a RISC-V architecture for an AI accelerator. I also build applied ML projects, including a lecture-to-notes pipeline and drug cardiotoxicity prediction models.
