Research /systems
$ Investigating AI infrastructure, LLM optimization, and intelligent systems. Publications, projects, and technical deep dives.
Projects
─1 activeOpen-source Rust and Python library that masks PII and secrets before they reach an LLM, in about 26 µs per 1 KiB chunk. Restores real values in replies with PrivacyGuard, and adds optional NER and LLM detection.
Publications
─8 entriesLLM Serving Fundamentals: Inference, Tokens, KV Cache and Latency
A practical, diagram-led guide to how a trained model becomes a fast, reliable API. Covers inference, tokens and context windows, prefill and decode, batch vs real-time workloads, throughput vs latency, the KV cache, quantization and streaming.
GGUF Memory Optimization
A systematic approach to estimating and optimizing VRAM usage for GGUF models in production environments.
Distributed Inference Systems
Design patterns for scalable LLM inference across heterogeneous GPU clusters with minimal latency.
KV Cache Quantization Strategies
Comparative analysis of quantization techniques for KV cache in transformer models and their impact on memory usage.
AI Agent Orchestration Frameworks
Evaluating orchestration patterns for multi-agent AI systems in enterprise workflows and decision-making.