Your day-to-day in this position

• Fine-tune and evaluate models for sales-conversation understanding, persona alignment, and domain-specific reasoning.
• Implement and benchmark instruction-tuned and adapter-based methods for context-efficient deployment.
• Build evaluation pipelines for model grounding, coherence, latency, and coaching-quality metrics.
• Design and maintain the multi-agent orchestration layer for pre-call, in-call, and post-call workflows.
• Build structured reasoning chains (prompt routing, tool selection, memory caching) using LangChain, LlamaIndex, or custom frameworks.
• Integrate streaming transcription, context compression, and stateful session memory for sub-second feedback loops.
• Lead the design of a low-latency distributed architecture (FastAPI, Redis, Pub/Sub, Cloud Run) to support concurrent real-time coaching sessions.
• Implement full observability for model behavior: latency tracing, token-level monitoring, and response analytics across agents.
• Establish model-versioning, evaluation dashboards, and automated rollback protocols.
• Prototype and test new prompting strategies, retrieval-augmented generation (RAG) configurations, and multi-modal fusion (speech, text, metadata).
• Collaborate with the data engineering team to design and manage conversational datasets for continual fine-tuning.
• Stay current with frontier research in LLM systems, reasoning agents, and real-time inference optimization.