Thesis is a fine-tuned environment for researchers
A secure AI workspace, that enables researchers, academics, and institutions to interact naturally with their own research data.
Train & Backpropagate
Forward and backward passes on your research corpus, computing and accumulating gradients within secure, dedicated environments.
Optimise Parameters
Gradients are applied to produce a domain-specific model that reflects your faculty’s research language and knowledge base.
Generate & Evaluate
The fine-tuned model is benchmarked and sampled to assess coherence, retrieval fidelity, and knowledge generalisation.
Checkpoint & Persist
Training states and resulting weights are securely saved for controlled deployment, future refinement, or continued training.
Supported Models
Qwen3
Llama 3
GPT-OSS
LoRA (Low-Rank Adaption)
LoRA fine-tunes large language models by training low-rank adapter layers while keeping the original weights frozen, enabling precise domain adaptation without retraining the full network.