UI/UX NGO Feedbacks: Enhance configurations and workflows

Summary

This issue tracks UI/UX feedback, workflow confusion, and feature improvement suggestions gathered from internal usage, NGO cohort discussions, and partner feedback sessions related to configurations, collections, evaluations, guardrails, and voice/audio workflows.

---

Configuration Creation Flow

• Improve UX for creating a new configuration from the Library page.
• Current flow involving:
  • loading an existing config
  • selecting + New Config from dropdown
    is non-intuitive and confusing.

Config ↔ Text Evaluation Navigation

• Improve navigation/UX between:
  • Configurations → Text Evaluations
  • Text Evaluations → Configurations
• Current screens expose too much information upfront, causing confusion:
  • Dataset / Create Dataset
  • Evaluations / Create Evaluation

---

Untitled Collection Placeholder

• Replace "Untitled Collection" with a better dummy/default naming mechanism.

Pending Collection Visibility

• While creating collections from existing documents:
  • show PENDING status immediately
  • prevent collection cards from disappearing until SUCCEEDED

Better Collection Creation Mechanics

• Improve UX for creating collections from documents.
• Current multiselect dropdown forces users to remember document names.
• Explore better interaction patterns:
  • drag-and-drop
  • searchable selection
  • document cards/list selection

---

Vector Store Visibility

• Move Vector Store ID/tool visibility to the primary view in Prompt Editor.
• Currently hidden behind the + Add Tool action, making discoverability poor.

---

Audio / Conversational AI UX & Features

Conversational Memory Layer

• Explore conversational memory layer for audio-heavy use cases.
• Need user-specific contextual memory persistence across multi-turn conversations.

Audio Cost Visibility

• Add predictable cost/token estimation for audio workflows.
• Current providers make token/credit usage difficult to estimate.

Multi-Provider Audio Evaluation Support

• Add support for multiple STT/TTS providers in evaluations UI:
  • Sarvam
  • ElevenLabs
  • Gemini
  • AI4Bharat
  • Bhashini

Better Provider-Specific Benchmarking

• Improve provider/model/language-specific benchmarking:
  • STT
  • TTS
  • Speech-to-Speech
  • RAG

The issue names no files, tests, or entry points. Start by splitting the feedback into separately scoped configuration, collection, prompt-editor, and audio workflow changes, then identify the relevant UI entry points for one agreed scope. Done should be defined per scope, since the issue currently combines UX changes, new capabilities, and provider support without acceptance criteria.