Voice-Based Livelihood Mapping & Skilling Recommendation Agent
Agents that start from unstructured input (voice, images, free-text documents) and turn it into structured action.
Background
Many rural workers cannot fill out long, text-based skilling-program forms, especially in English. A natural voice conversation in their own language would remove this barrier to accessing skilling programs.
Objective
Develop a voice-first agent that converses with a person about their current work and skills in their own language, maps this to structured skill categories, and recommends relevant government skilling programs.
Key Features
1. Voice input in at least one Indian language 2. Natural conversational flow to understand occupation and experience 3. Mapping of free-text responses to structured skill/NSQF categories 4. Skilling program recommendation with reasoning
Expected Solution
A speech-to-text pipeline for voice input A conversational agent that asks natural follow-up questions A skill-mapping layer (free text → structured categories) A recommendation engine matching mapped skills to programs
Suggested Technology Stack
Speech-to-text: Whisper (open-source) / Google Speech-to-Text Conversation: LLM-based dialogue agent Mapping: Embedding similarity or LLM classification Data: Simplified NSQF category list (20–30 categories), 15–20 curated skilling programs Text-to-speech: (optional) For a full voice loop
Expected Outcomes
Improved accessibility of skilling programs for non-literate or non-English-speaking users Higher enrollment in relevant, well-matched training programs A reusable voice-first intake pattern for other government services
Possible Use Cases
A rural worker discovering a relevant Skill India program by voice A field NGO worker helping community members enroll A basic IVR-style skilling helpline
Evaluation Parameters
Transcription accuracy Naturalness and effectiveness of the conversation Accuracy of skill-category mapping Relevance of final recommendations