Cortance
A remote developer network platform that records developer profiles and project information for talent matching and project statistics.
Timeline (last 90 days)
Cortance reviewed 69 AI product projects in its developer network, of which 42 were built on existing models and 27 trained or fine-tuned models.
Among the 69 AI products, computer vision and LLM features within existing products are tied as the largest categories, with 11 products each.
Language model-related work accounts for 37 of the 69 products, and very little of it involves training.
In the sample, 10 products used retrieval, 5 fine-tuned or custom-trained language models, and 2 products did both.
In the sample, 7 products use AI to change how software is built, rather than as end-user-facing features.
In the sample, 10 products trained or fine-tuned models for a single task, covering player behavior prediction for a game studio, oilfield DNA classification, predictive maintenance for factory systems, and agent research that learns by playing games.
The defense effort forms an independent cluster (6 products), of which 2 products solve the navigation problem when GPS is jammed or missing, and 4 handle onboard detection, tracking and guidance for drones.
Agent projects share one design habit: the agent stops before doing anything expensive; 6 products described allowlisted tools, human approval queues, or confidence thresholds to pause the workflow.
All Facts (8)
Cortance reviewed 69 AI product projects in its developer network, of which 42 were built on existing models and 27 trained or fine-tuned models.
50%UnverifiedAmong the 69 AI products, computer vision and LLM features within existing products are tied as the largest categories, with 11 products each.
50%UnverifiedLanguage model-related work accounts for 37 of the 69 products, and very little of it involves training.
50%UnverifiedIn the sample, 10 products used retrieval, 5 fine-tuned or custom-trained language models, and 2 products did both.
50%UnverifiedIn the sample, 7 products use AI to change how software is built, rather than as end-user-facing features.
50%UnverifiedIn the sample, 10 products trained or fine-tuned models for a single task, covering player behavior prediction for a game studio, oilfield DNA classification, predictive maintenance for factory systems, and agent research that learns by playing games.
50%UnverifiedThe defense effort forms an independent cluster (6 products), of which 2 products solve the navigation problem when GPS is jammed or missing, and 4 handle onboard detection, tracking and guidance for drones.
50%UnverifiedAgent projects share one design habit: the agent stops before doing anything expensive; 6 products described allowlisted tools, human approval queues, or confidence thresholds to pause the workflow.
50%