LLM serving, multimodal assistants, fine-tuning and agents. Each one comes with the numbers that show whether it worked.
3.1×
token throughput
LLM infrastructure · 2025
An inference backend serving a LLaMA-3 8B chat model with vLLM behind a FastAPI microservice, tuned for throughput on a single 24 GB GPU.
<400 ms
response latency
LLM + RAG + speech · 2024 – Present
A multimodal LLM assistant that listens, reads emotional tone from the voice, retrieves context and answers in speech, text or images.
96%
accuracy, 5 classes
Fine-tuned LLMs · 2024
LoRA fine-tunes of LLaMA-3 8B and Gemma-2B that classify news articles across five political-bias categories.
−32%
tool calls per task
Multi-agent systems · 2025
LLM-based Planner, Researcher and Executor agents that coordinate to research, plan and carry out multi-step tasks with real tools.