Bona Jasu Aishwarya

• Architected an agentic RAG (Retrieval-Augmented Generation) pipeline using a 'Router' pattern to autonomously determine if infrastructure queries require complex SQL generation for metrics or semantic search for patterns.
• Implemented vector embeddings and semantic search using specialized libraries to improve the relevance of retrieved technical context by 15%, significantly reducing LLM "hallucinations."
• Designed a Gemini-powered SQL Agent that translates natural language into optimized PostgreSQL queries, allowing stakeholders to reason over 100 GB+ metrics tables without manual SQL intervention.
• Engineered a Predictive Utilization Engine that forecasts pctIdle metrics to identify impending high-utilization states, proactively preventing server crashes across critical infrastructure.
• Optimized XGBoost forecasting models by implementing an automated Optuna-based tuning pipeline, achieving the high precision required to reliably predict system failures and capacity bottlenecks.
• Engineered an automated observability bridge using APScheduler to poll Prometheus metrics across 50+ namespaces every 15 minutes, ingesting over 25 unique metric types (CPU, Memory, Throttling) into PostgreSQL and Splunk.
• Optimized large-scale data ingestion by implementing table partitioning, aggregating, and batch processing using psycopg2, ensuring the LLM only queries relevant data slices rather than full table scans.

Key Projects:

Agentic Observability Platform:

• Developed an AI-driven bridge for the Data Lakehouse Platform metrics (Iomete) that utilizes Gemini 1.5 Flash to provide instant, natural language insights into infrastructure spikes.
• Built an agentic RAG platform on top of high-velocity metrics tables, allowing users to query live infrastructure health using Gemini 1.5 Flash.

Predictive Infrastructure Resilience Engine

• Engineered a time-series forecasting system to monitor host health by predicting CPU Idle percentages (pctIdle).
• Utilized XGBoost and Optuna to reach optimal model accuracy, allowing the system to flag "crash-risk" servers before utilization peaks, supporting long-term resource planning, and system stability.

Catalog Migration Tool:

• Engineered an automated Catalog Migration Tool for Apache Iceberg, facilitating seamless data movement across IOMETE catalogs, while maintaining 100% data integrity through automated Avro/JSON metadata path rewriting.

MDP Observability Integration:

• Developed a Python-based ETL bridge between technical logs (Splunk) and business intelligence tools.
• Developed a time-series forecasting engine to provide long-term capacity predictions for infrastructure metrics, resulting in a dataset used for year-end resource planning.