KE Y RE SPONSIBI L ITI E S
LLM Hosting & Serving
RAG Pipeline Engineering
Agentic AI Systems
Kubernetes & GPU Infrastructure
Open-Source Stack Stewardship
Deploy and maintain Ollama (llama.cpp backend)
as the primary inference engine on Sybyl GPU
Server
›
Manage GGUF-format model quantization
(Q4_K_M, Q8_0) for efficient inference —
optimising for throughput, latency, and VRAM
efficiency
›
Tune inference parameters: context length, GPU
memory fraction, batch size for throughput and
latency targets
›
Monitor GPU utilisation, VRAM allocation, and
inference performance; resolve bottlenecks
›
Evaluate alternative open-source engines (vLLM,
llama.cpp standalone, LMDeploy, TGI) for future
hardware upgrades
›
Manage model registry and version control for
hosted models across the Qwen3, Mistral, and
LLaMA families
›
Design, implement, and maintain production
RAG pipelines using LangChain and LlamaIndex
›
Manage and tune vector stores (ChromaDB,
FAISS) for document retrieval at scale
›
Integrate web search via SearXNG into retrieval
pipelines
›
Implement chunking strategies, embedding
model selection, hybrid search (dense + BM25),
and re-ranking
›
Evaluate and improve retrieval accuracy, latency,
and hallucination rates across pipelines
›
Deploy and maintain Hermes (Nous Research,
nousresearch/hermes-agent) on CPU worker
nodes, connected to Ollama via OpenAIcompatible
API
›
Deploy and integrate OpenClaw agentic
framework within the stack
›
Build multi-agent pipelines: tool-use, function
calling, agent chaining, multi-step reasoning
›
Debug agentic runtime behaviour, tune
prompting strategies for reliability and latency
›
Manage GPU workloads on Kubernetes: node
taints/tolerations, GPU resource requests,
namespace management
›
Work with NVIDIA GPU Operator, CUDA 12.x
drivers, and GPU monitoring tooling
›
Author and maintain Kubernetes manifests and
Helm charts for all AI services
›
Contribute to cluster reliability: Longhorn
storage, Traefik + MetalLB ingress, networking
›
Maintain the Ubuntu 22.04 LTS + NVIDIA driver
550/560 environment on GPU nodes
›
Maintain LibreChat (MIT) as the production AI
front-end and SearXNG for search
augmentation
›
Evaluate new open-source tools and models;
produce written integration recommendations
›
Contribute upstream to open-source projects
where appropriate