Shreyas Dasari

Delivered 2× faster research planning outputs by designing a hierarchical reasoning agent in Python that decomposed prompts into structured sub-tasks for faster multi-step decision support. Reduced literature review effort by 70% via ResearchPal, a conversational assistant that retrieves, ranks, and summarizes papers (including arXiv ingestion) using embeddings and transformer-based summarization. Built end-to-end data pipelines to ingest inputs, clean and validate data, and generate structured outputs for downstream analytics. Increased product engagement by 45% by shipping interactive analytics-style AI apps (Streamlit + Python) for actionable insights and scenario simulations. Collaborated with product and research stakeholders to translate ambiguous requirements into measurable KPIs, experiment metrics, and interpretable outputs.

Improved AI chatbot accuracy and engagement by 40% by integrating Azure AI Search with chat completions and semantic result parsing in JavaScript. Designed and evaluated a BERT-based semantic similarity model to match founder profiles, increasing match quality by 35% and reducing false positives. Built SQL-backed data retrieval logic to support profile comparison, ranking, and analytics validation. Documented and productionized an end-to-end Azure AI Search pipeline (data ingestion, indexing, query patterns), accelerating onboarding and reducing iteration time for new engineers. Collaborated with cross-functional teams to translate business goals into measurable model performance metrics and user-facing analytics.

Delivered 98% accurate outputs across 78 monthly data requests by analyzing operational datasets, validating inputs, and standardizing SQL-based reporting logic. Reduced processing time by 40% by automating repeatable data cleaning, validation, and reporting workflows. Led global incident triage and root-cause analysis using operational metrics and trend analysis, reducing downtime by 70% and preventing $500K+ in business losses. Executed high-pressure production issue resolution by prioritizing data signals, coordinating stakeholders, and delivering clear, data-backed remediation plans within 3 hours.