Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{200660,
        author = {Viren Jalamkar and Tushar Uddhage and Litesh Patil and Pooja Chaudhari},
        title = {MediCall: A Real-Time Hinglish Voice AI Receptionist for Autonomous Hospital Appointment Management Using LLM Tool-Calling},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {12},
        number = {12},
        pages = {2140-2144},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=200660},
        abstract = {This paper presents MediCall, an end-to-end production-deployed AI voice receptionist built for the Indian outpatient healthcare context. The system embodies an AI persona named Priya, powered by Meta’s LLaMA 3.3 70B large language model served at 276 tokens/second via Groq’s Language Processing Unit (LPU) — the fastest independently benchmarked provider for 70B models. MediCall integrates Twilio Programmable Voice PSTN telephony to handle inbound patient calls in natural Hinglish (Hindi-English code-switched) speech, employs structured JSON LLM tool-calling against a Firebase Realtime database for live appointment management, and autonomously completes booking, rescheduling, and cancellation without human intervention, 24×7. A Python FastAPI backend enforces all scheduling policies deterministically. A React.js administrative dashboard provides live slot management, manual override, and attendance tracking. Deployed on AWS EC2 free-tier, the system achieved 91.1% end-to-end task completion across 45 interactions, with standard latency averaging 1.15 s. Key contributions: (1) two-phase LLM tool-calling pipeline eliminating appointment hallucination by design; (2) systematic Hinglish spoken time-format localisation; (3) task-specific least-privilege safety guardrail architecture; (4) empirical proof that production-grade AI reception is achievable at near-zero infrastructure cost.},
        keywords = {LLM Tool-Calling; Hinglish NLP; Hospital Appointment Scheduling; Twilio VOIP; Groq LPU; LLaMA 3.3 70B; FastAPI; Firebase; React.js; Voice AI; AI Guardrails.},
        month = {May},
        }