All work
LLM & Agents · Enghouse Interactive · 2024 — Present
Best-Value LLM Benchmarking
A benchmarking and evaluation programme that matches every LLM use case with the most cost-effective model — closed or open source, cloud or local.
01 The problem
Teams defaulted to expensive frontier models for every task, with no objective way to tell whether a cheaper model would do the job just as well.
02 What I did
- Built a repeatable benchmark across GPT, Gemini, Llama, Mistral and DeepSeek on Vertex AI, Azure, AWS Bedrock and locally via Ollama and vLLM.
- Paired it with an automated evaluation framework that scores LLM output quality per use case.
- Mapped each production use case to the best-value model and hosting option — using open-source models with data masking where data was sensitive.
03 Impact
- 50–60% reduction in model costs across several use cases
- Objective, repeatable model selection instead of guesswork
- A clear path to local / open models for data-sensitive workloads
Client work — details generalised to respect confidentiality. Happy to discuss specifics in an interview.