Ankur Singh

At Techificial.ai, I architect the conversational layer of a prompt-based agentic voice platform, mapping advisor workflows into intent detection, dialogue orchestration, and multi-step reasoning. I designed its prompt architecture with context injection, tool-calling patterns, guardrails, output validation, and graceful fallbacks.

I built stateful call sessions with isolated rooms, JWT-secured FastAPI, and PostgreSQL memory that reloads across calls. I integrated OpenAI LLMs with Deepgram STT and TTS over LiveKit for real-time audio, and created conversation quality metrics and evaluation datasets from interaction logs.

My projects include voice agents for education advising, insurance guidance, and general assistance, as well as LearnSprint, an AI-powered active learning platform supported by a BMU Seed Grant and presented at HCI International 2026. I’m interested in voice-first AI agents, thoughtful UI/UX, and human-centered health technology at the intersection of biosignals and machine learning.