Model developer

What you will do

You will research, design and train Harpe's proprietary models from scratch: models that decide by what a program does, language models that reason over documents and data and cite their sources, vision models on geospatial imagery and whatever comes next.

It is research and development in one role: not fine-tuning someone else's model, but making the architecture, the loss function, the data and the evaluation yours and taking them all the way to a model that runs locally, on the client's machine or servers, with little memory, little power and offline.

Harpe's models are implemented in C; Python is for exploring and prototyping.

Who we are looking for

We are looking for a research profile with a background in mathematics or physics, a PhD or equivalent research experience, solid foundations in linear algebra, probability and optimisation, publications or original work in statistical learning and the ability to take an idea from paper to C code.

Anomaly detection, time series, language models, computer vision and low-resource models are a plus.