Rohith Sadanala

About

LLM/agent engineer who has shipped production RAG chatbots in sustainability-focused domains, including a packaging recommendation assistant that standardized messy user inputs and used Pinecone-backed retrieval over product/regulatory data. Experienced orchestrating end-to-end ML workflows with Airflow and AWS Step Functions/Lambda, emphasizing reliability (property-based testing, circuit breakers, OpenTelemetry) and measurable performance (latency/cost). Partnered closely with non-technical leadership to ship 3 weeks early, driving adoption by 150+ businesses and ~20% reported waste reduction.

Key Strengths

• Built and deployed production LLM chatbot for sustainable packaging recommendations
• Improved intent accuracy from 62% to 88% using few-shot prompting (5 curated I/O examples)
• Designed preprocessing to handle inconsistent/vague user inputs (unit normalization, entity extraction, numeric-range mapping) to reduce hallucinations
• Implemented RAG with Pinecone retrieval (top-3 docs) over product/regulatory knowledge base
• Orchestrated ML/LLM workflows with Apache Airflow (DAGs for API integration, embeddings, indexing, backfills)
• Used AWS Step Functions + Lambda for production coordination with retries/exponential backoff and parallel fan-out; reduced end-to-end latency by 40%
• Strong reliability/testing discipline: unit + property-based (Hypothesis) + Dockerized integration tests; circuit breakers/fallbacks; OpenTelemetry monitoring
• Metrics-driven evaluation and model/retrieval selection (success rate, P95 latency, cost/query; BM25 vs dense; A/B testing)
• Delivered cross-functionally with non-technical stakeholder via weekly demos and dashboarding; shipped 3 weeks early
• Demonstrated business impact: 150+ businesses adopted recommendations; reported 20% waste reduction

Experience

Education

Skills

Languages