Fine-tuned model active: All analyses are using your custom-trained model
Semantic Cache
β
entries
Admin Overrides
β
never expire
RAG Embeddings
β
event vectors
Training Examples
β
collected
Pending Fine-tune
β
β₯4β ready
Cache Hits (24h)
β
LLM calls saved
β Rate an Analysis β builds fine-tuning training data
βΌ
After reviewing a Full Analysis on the dashboard, paste it here and rate it.
Ratings of 4β5β are automatically saved as fine-tuning training examples.
Correct how a specific headline is classified. Writes to the semantic cache immediately β the same
headline will always use your correction going forward.
π Training Examples
βΌ
Date
Location / Source
Rating
Preview
Status
Loadingβ¦
π§ Fine-tuning Pipeline
βΌ
How it works
Rate analyses 4β5β to build up training examples.
Once you have β₯10 high-quality examples, submit a fine-tune job to OpenAI.
OpenAI trains a custom gpt-4o-mini on your African security data (15β60 min).
Once complete, all future analyses automatically use your fine-tuned model.
The model learns Africa Watch's style, quality standards, and regional context over time.
π RAG Explorer β inspect historical event retrieval
βΌ
Test what historical events are retrieved for a query. These are injected into analysis prompts as context.