Satish Satish

A production-grade GenAI Financial Agent for GE Vernova, enabling natural language queries across thousands of legacy documents. As an AI engineer, I owned the serverless RAG pipeline using Python, AWS Lambda, FastAPI, and Azure OpenAI GPT-4.

My Core Contributions:
• Multi-Format Pipeline: Built a zero-downtime serverless Lambda pipeline supporting 12+ file formats (PDF, DOC/X, PPTX, XLSX/S/B, images) via Textract OCR.
• Lambda Optimization: Engineered a zero-dependency OLE2 parser for Lambda, bypassing a 300MB LibreOffice binary constraint to recover legacy data.
• Table Extraction: Parsed binary .doc markers using olefile to generate structured markdown tables inside an isolated runtime.
• Hybrid Retrieval: Combined 3072-dim pgvector embeddings, tsvector text search, RRF ranking, and Cohere reranker to hit a 3.88/5 relevance score.
• CI/CD Quality Gates: Automated evaluation runs on deployment using RAGAS to enforce KPI thresholds (Correctness, Groundedness) and eliminate regression.
• SME-Validated Impact: Secured a 5.0/5 domain accuracy rating from Subject Matter Experts for answer correctness and completeness.

Served as the lead data engineer for the development of a "Golden Record" audience platform designed to unify listener data across multiple fragmented sources.

My primary responsibility was architecting a Medallion data architecture (bronze, silver, and gold layers) using Databricks and PySpark on AWS. I engineered high-performance ETL pipelines that processed millions of records daily, ensuring that disparate data points from streaming services and social platforms were resolved into a single, high-fidelity user identity. To ensure production-grade reliability, I automated the infrastructure deployment using Terraform and integrated AWS Lambda for event-driven processing. I also implemented a robust data quality framework that performed automated validation at the silver layer, reducing data errors by over 20%.

This project directly enabled the client to perform advanced predictive modeling and hyper-personalized marketing, resulting in a significant uplift in listener engagement and data-driven revenue growth.

Developed a comprehensive data analytics platform for Maruti Suzuki India Limited (MSIL) to optimize dealership supply chain operations. The project focused on parts back-order analysis, enabling the product analysis and planning (PAP) team to identify the root causes of inventory shortages and transition from reactive to proactive inventory management.

My role involved designing and implementing a serverless ETL architecture using Amazon S3, AWS Glue (PySpark), and Amazon Redshift. I also integrated Amazon Athena to enable rapid ad hoc validation of large datasets. To bridge the gap between data and action, I built a secure API layer using AWS API Gateway and Lambda, which exposed curated Redshift data to real-time executive dashboards and dealership applications.

This solution significantly accelerated the insight-to-decision cycle, ensuring critical parts were available when needed and reducing operational bottlenecks across the dealership network.

Designed and implemented a scalable Snowflake-based data warehouse to consolidate multi-source data into a unified analytics platform. The solution ingested data from S3, APIs, and operational systems, loading raw data into Bronze layers and transforming it into curated Silver and Gold datasets using dbt.

I developed modular dbt models with incremental processing to optimize performance and ensure efficient data transformations. I implemented a robust data quality framework combining dbt tests, SQL validations, and Python-based automation to enforce completeness, accuracy, and referential integrity across pipelines.

My work included orchestrating end-to-end workflows using Airflow, enabling automated ingestion, transformation, testing, and monitoring. I integrated CI/CD pipelines for version control and automated deployments, ensuring reliable and maintainable production workflows. I also built monitoring and alerting mechanisms to track pipeline health, detect anomalies, and prevent data issues from propagating downstream.

Finally, I delivered trusted, analytics-ready datasets that improved reporting efficiency, reduced data quality issues, and enabled faster, data-driven decision-making across business teams.