ai research and development
systems that think, create, and evolve.
an independent ai research and development lab — building agentic frameworks, studying frontier models, and engineering systems at the intersection of performance and intelligence.
explore
journal
latest research
openclaw and the agentic OS thesis
exploring why nvidia's jensen huang called openclaw 'the next chatgpt' — and what an open-source agentic operating system means for the future of personal computing.
bitnet and the death of floating point
microsoft's 1-bit LLMs run 100B parameters on a single CPU — what ternary weights mean for the future of edge inference and the end of floating point dominance.
meta's avocado problem
codename avocado delays, gemini licensing talks, and what meta's stumbles reveal about the open vs closed AI strategy debate.
nvidia GTC 2026 — the trillion dollar signal
jensen huang's $1T demand projection, vera rubin's 10x perf/watt, feynman architecture teased for 2028, and the enterprise agent play with nemoclaw.
research
active studies
openclaw enterprise evaluation
studying nemoclaw/openshell enterprise deployment patterns. privacy router architecture, local-first inference with nemotron models, sandbox security model.
1-bit inference benchmarks
benchmarking bitnet b1.58 models on various hardware. ARM vs x86 CPU performance. energy consumption profiling against FP16/BF16 baselines.
structured compression with openzl
evaluating meta's openzl format-aware compression framework for AI workloads. 2x compression ratios at 200+ MB/s. testing on model weights and training data pipelines.
zero-knowledge credential systems
studying google's longfellow ZK library for privacy-preserving identity verification. ECDSA-based ZK proofs, EU eIDAS compliance, agentic authentication applications.
engineering
projects & tools
ideonexus
an agentic research framework for managing AI/ML projects — tracking ideas with genealogy, designing sandboxed experiments, and monitoring the competitive landscape.
saengil.ai
this website — an independent AI research and development portal. static site architecture, deployed on cloudflare, designed as a living document of ongoing work.
model evaluation suite
a benchmarking toolkit for evaluating frontier language models across reasoning, code generation, and agentic task completion. standardized evaluation harnesses and reporting.
metrics