Piyush Kautkar

About

CMU candidate with hands-on experience taking LLM concepts from research prototypes toward production-ready designs (structured outputs, guardrails, failure-scenario evaluation). Also partnered with sales/customer teams at Mazecare to drive adoption with Dontia Alliance (largest dental clinic chain in Singapore) and engaged Singapore government stakeholders, bridging clinical workflow needs with IT security/integration concerns.

Experience

Education

Key Strengths

• Pragmatic LLM system design (deterministic rules + LLM where ambiguity adds value)
• Production-readiness focus: guardrails, structured outputs, and safe fallback behavior
• Strong real-time debugging of agentic systems using logs/observability to isolate failure points
• Designing explicit contracts between reasoning and execution to stabilize workflows
• Demo/workshop effectiveness for developers (hands-on, failure modes, reusable snippets)
• Cross-functional deal support: tailoring messaging for clinicians vs IT to unblock adoption
• Owned end-to-end delivery of production ChatOps chatbot (scope → build → launch → iterate)
• Strong LLM safety/guardrails mindset (read-only launch, tool allowlisting, strict schemas, citations, fallback behaviors)
• Metrics-driven iteration improving trust and reducing follow-up questions via structured outputs (summary/data/assumptions)
• Performance and cost optimization through caching and minimizing tool calls
• Production RAG system design with relevance thresholds, retrieval caps, and sensitive-data-aware logging
• Built automated LLM eval/regression loop with human review and deployment blocking on accuracy drops
• Improved time-based query accuracy by adding relative-date resolution preprocessing
• Designed scalable, observable ingestion pipeline for messy sensor data with validation and quarantine handling