Bhargav Tavethiya

• • Led backend development for real-time AI voice agents — scalable Python services for inbound and out bound calling over Twilio and Vonage, with live transcription, LLM response generation, and real-time TTS playback.

• • Built and optimized low-latency voice pipelines handling STT, LLM reasoning, TTS, interruption handling, and call-state management; designed deterministic graph-traversal logic to reliably guide conversation flow.

• • Designed a voice-to-voice simulator that runs conversations between AI voice agents and simulated cus

tomers, enabling hundreds of automated simulations to test and validate agent behavior at scale.

• • Deployed custom LLMs (Gemma, Llama) on Cerebrium using vLLM and Ollama, and implemented audio

caching to cut token usage and operating cost.

• • Single-handedly transitioned production deployment from Cerebrium to Azure without compromising qual ity or reliability.

• • Engineered the critical MVA module from scratch and integrated multiple AI providers (OpenAI, Anthropic, Deepgram, ElevenLabs, Cartesia, Azure AI) into a modular, provider-swappable pipeline.

• • Profiled and optimized end-to-end voice latency (WebSocket setup, STT/TTS connections, graph traversal, first-audio delivery), and built end-to-end observability with Grafana and Sentry to surface and report issues, speeding incident diagnosis.

• • Built a collaborative project-management tool with LLM-based file querying (OpenAI, Claude) using RAG; enabled content search across formats (video, audio, documents), contributing to data ingestion and hybrid search.

• • Developed the Adapt Platform, letting users upload and query their own data with pgvector-backed retrieval and GPT-powered real-time summarization and prompt optimization.

• • Worked on deploying services to AWS using ECS and ECR.

• • Contributed significantly to MedWriter (medwriter.ai), an LLM-powered healthcare documentation solution

using GPT and Claude; worked on the live transcription, templatizing, and summarization flow to auto

generate clinical and administrative documents, streamlining doctors' tasks and reducing patient processing time by 40%.

• • Contributed significantly to the backend and AI components of KompetencyAI, an AI-powered competency assessment platform for regulated, high-stakes environments (healthcare, pharma, biotech) — a joint ven ture with SeerPharma.

• • Contributed to the agentic, multi-agent architecture where AI agents take on distinct roles (clients, col

leagues, supervisors) to drive realistic, real-time conversational role-play that tests decision-making, critical

thinking, and problem-solving.

• • Contributed significantly to a configurable rubric engine and automated grading pipeline delivering instant, objective, bias-free feedback, plus reporting and analytics surfacing performance trends and skill gaps, with

quality controls including approval flows, version control, and role-based permissions.