Resume.

Summary

ML engineer building on-device LLM inference, PyTorch LLM training and post-training, and GPU and AI accelerator benchmarks.

Technical Skills

Training & Post-Training: PyTorch, Hugging Face Transformers, PEFT/LoRA, TRL, DPO, GRPO, torch.compile

Inference & Optimization: ExecuTorch, llama.cpp, vLLM, ONNX Runtime, MLX, GGUF and GPTQ quantization

Accelerators: CUDA (H100, H200), ROCm (MI300X), AWS Trainium/Inferentia, Google TPU, mobile NPUs

Software & Cloud: Python, Kotlin, TypeScript, SQL, Docker, Kubernetes, Terraform, AWS, Google Cloud

Professional Experience

Experimental Machines, Machine Learning Engineer

Aug 2026 – Present

• Shipped OpenWeights to Google Play; its ExecuTorch LFM2.5 1.2B decodes 2.3× faster than llama.cpp on Tensor G5.

• Published 16 ExecuTorch LLM exports to Hugging Face for phone CPUs, GPUs, and Qualcomm and MediaTek NPUs.

• Built ExecuServe, an OpenAI- and Anthropic-compatible server that runs ExecuTorch LLMs from an Android phone.

• Merged a fix into PyTorch ExecuTorch that stops LFM2 models from carrying state into a new conversation.

• Fixed RoPE in MediaTek's stock NPU export for Qwen and Llama, lifting Qwen3 top-1 match with Hugging Face to 100%.

• Secured a $2,000 compute grant from Lambda for the team's GPU training and benchmarking work.

ViableView, Senior Data Engineer

Feb 2026 – Present

• Designed an email-enrichment pipeline processing 2.6M records at 240K/day and achieving 0.94 macro-F1 on validation.

• Built ingestion for 2.7M registry records at 200K/day and an autoscaled multi-cloud pipeline validating 800K at 80K/day.

Kollab, Associate Cloud Engineer

Sep 2025 – Feb 2026

• Delivered a patient-search AI system for a Philippine hospital in 30 days using Google Cloud, Vertex AI, and AlloyDB.

Projects & Research

Small LLM Trained from Scratch to RLVR

Aug 2026

• Pre-trained a 151M-parameter LLM (GQA, RoPE, SwiGLU) on 8.26B tokens, part of a 20-hour full pipeline on one H100.

• Implemented SFT, DPO, and GRPO in plain PyTorch; GSM8K answer format went from 5.5% after SFT to 98.5% after GRPO.

LLM Fine-Tuning and Serving on AWS Trainium and Inferentia

Aug 2026

• LoRA fine-tuned Llama 3.1 8B on Trainium at 68.3% MFU (2,952 tok/s) and served the merged model with vLLM on Inferentia2.

• Matched CPU labels on all 10,000 ImageNet images for compiled CLIP and SigLIP, and caught a CLIP graph returning NaN.

MI300X vs. H200 GPU Benchmark

Jul 2026

• Measured 164 vLLM serving points; H200 ran 8B training steps 1.32–1.40× faster despite MI300X's 1.32× FLOPs lead.

• Showed MI300X winning 70B FP8 serving at 128–256 concurrency by up to 1.14× and fitting 70B in BF16 where H200 cannot.

Visual-Qwen: Multimodal Short-Form Video Classification

Dec 2025

• Built a tri-modal EVA-CLIP/Whisper/Qwen3-4B VLM detecting “sludge” videos at 97.19% F1 on 300 held-out videos.

• Published the model, code, and a 2,000-video multimodal dataset spanning TikTok and YouTube Shorts.

Speaking & Community

International Conference Speaker, PyTorch Foundation & GitHub

Apr 2026 – Dec 2026

• Presenting “When the KV Cache Cannot Roll Back: Prefix Reuse on Android” at PyTorch Day Japan, Tokyo.

• Presenting “When Is Local Copilot Worth It?” at GitHub Universe, San Francisco.

• Presented “De-mystifying PyTorch for ASICs” at PyTorch Conference Europe, Paris.

PyTorch Ambassador, GitHub Campus Expert, AWS Community Builder

Dec 2024 – Present

• Selected as a 2026 PyTorch Foundation Ambassador to run events, create content, and mentor PyTorch users in the Philippines.

• Trained 4,000+ students and professionals in AI/ML, software engineering, and data engineering across 25+ talks.

• Secured $30,000+ in sponsorships from GitHub, Tavily, and Vercel to support technology education across Luzon.

Certifications

• PyTorch Certified Associate (PTCA)

• Google Cloud Professional Machine Learning Engineer

• AWS Certified Machine Learning Engineer – Associate

• Google Cloud Associate Cloud Engineer

Education

BS in Computer Science with specialization in Software Engineering, FEU Institute of Technology

BS

Aug 2022 – Sep 2026

4-Year President's Scholar with High Honors; Leadership Award (3 of 921 graduates); CS Academic Varsity

Awards & Recognition

AWS re:Invent 2025 All Builders Welcome Grantee, AWS: sole Filipino among 340 international recipients

Dec 2025

PH100 Honoree, StellarPH

Sep 2025

TESDA Academe Expert: validated two Philippine national AI and data competency qualifications

Aug 2025

National AI Prompt Design Challenge Champion, Straits Interactive

Jul 2025