Platform Overview
console.kaapi.ai is the interface for your product, engineering, and programme teams. It enables you to build and operate AI features through the following sections.
Documents
Upload knowledge-base files. Optionally convert to clean markdown for better retrieval — useful for scanned content, handwritten Indic-language text, and complex tables.
Collections
Create provider-managed vector stores from your uploaded documents. Reference them in any AI configuration.
AI Configs
Build the recipe — provider, model, parameters, system prompt, vector store, guardrails. Version it. Lock it for production.
Quick Test
Throw sample inputs at an AI config in a built-in chat UI. Sanity-check responses before running formal evals.
Datasets
Upload golden datasets (CSV) for evaluation, or submission datasets for batch assessment runs. Map columns to text inputs, file attachments, and ground-truth scores.
Evaluations
Batch-run a golden dataset against one or more AI configs. Get LLM-as-judge scores, semantic similarity scores, and side-by-side comparison. Use a duplicate factor to test reproducibility.
Assessments
Run rubric-based scoring on multimodal submissions. Compare up to four AI configs on the same dataset. Structured output: per-metric scores, reasoning, attachment summary, assessee-facing feedback.
Guardrails
Configure input and output validators per AI config — slur lists, ban lists, PII detection, topic relevance, toxicity, gender-bias removal.
Chains
Compose pre-built pipelines like speech-to-speech with file search. Single endpoint, intermediate callbacks, evals support.
Cost & Usage
Track token spend and request volume per provider, per model, per AI config.
User Management
Add team members to your organisation and collaborate on AI configs and runs.
Some features (e.g. AI mentoring, FAQ generation, advanced eval analytics) are on the roadmap and rolling out progressively. Check release notes for the latest.
Next: see how these pieces come together in the Core Use Cases.