Detroit, Michigan, United States
Hi, am Saurabh. I work on the messy middle of data and AI problems the part where the pipeline is broken, the requirements are fuzzy, and someone needs a defensible answer before the next standup. Data at scale tends to break in predictable ways, and I've built my work around catching it before it does. I hold an MS in Data Science & AI from the University of Michigan (4.0 GPA). Before grad school, I spent four years at Thales building data pipelines for Digi-Yatra, the biometric boarding system used across Indian airports where I learned how much "AI/ML" actually depends on the unglamorous work underneath it. A few things from the last year I'm proud of: At Great American Insurance Group, I built an Azure AI Foundry RAG pipeline over UC4 job documentation that cut developer lookup time by ~70%. In the same internship, I resolved production defects in a Snowflake migration and cut ETL memory usage 50% with a Bash rewrite. At Accounting Aid Society, I worked directly with 200+ clients on federal, state, and city tax filings, using SQL and QuickBase to reconcile high-volume financial data and tighten reporting accuracy ahead of audit. At Sera AI, I built a CBT-grounded mental health chatbot on LangGraph, chaining Groq LLMs (Llama, GPT) with an Astra DB vector store then tested it with 20+ students and used their feedback to fine-tune Llama and Gemma. The tools I reach for most: Python, SQL, PySpark, LangChain/LangGraph, Azure AI Foundry, Snowflake, Power BI, AWS, Docker, FastAPI. I also write on Medium about SQL joins, window functions, and hypothesis testing usually when I want to force myself to think something through properly. My measure of success isn't lines of pipeline code. It's whether the data can be trusted, whether the system holds up under pressure, and whether what I build today still works as things scale. Open to full-time roles in Data Analytics, Data Science, AI/ML Engineering, and Data Engineering. If you're hiring or just want to talk about RAG, agentic workflows, or data infra I'd love to chat. Email: [email protected] Portfolio: https://sauraoworks.netlify.app/
• Analyzed 200K+ global sales records across three years in SAP Databricks to identify revenue drivers, seasonal trends, and underperforming segments, informing proactive business planning. • Built demand and revenue forecasts using SARIMA, Holt-Winters, and XGBoost, achieving 4% MAPE via backtesting at 80% confidence intervals to drive inventory optimization and revenue strategy. • Developed agentic AI pipelines for data ingestion, model retraining, and forecast generation, automating the end-to-end forecasting workflow.
• Collaborated with Accounting, Finance, and Compliance teams to resolve data issues and prepare governmental reports, ensuring audit readiness and data governance compliance. • Analyzed and reconciled high-volume financial and tax data using SQL, ERP QuickBase, Machine Learning, and Python reducing calculation discrepancies and significantly improving reporting accuracy. • Served 200+ clients directly, gathering financial information and guiding timely, accurate filings of federal, state, and city income tax returns.
•Developed a mental health chatbot using Cognitive Behavioural Therapy (CBT) principles that guides users through a step-by-step process for addressing mental health issues utilizing real world knowledge base. •Built a persistent Lang-Graph workflow that chains multiple Groq LLM including Llama, Open AI calls with real-time data from Astra DB vector database- Beck's Cognitive Model all within an AI agent router framework. •Deployed a MVP version on Railway-GitHub, tested it with 20+ college students and peers, and used the collected data from PostgreSQL to fine-tune the small language model.
• Built a production RAG pipeline in Microsoft Azure AI Foundry vectorized UC4 job documentation with OpenAI text-embedding-ada-002, implemented KNN-based hybrid search indexing, and deployed via Render reducing developer manual lookup time by 70%. • Resolved real-time production defects during a Snowflake data migration by engineering a Linux Bash optimization for UC4-Automic ETL jobs, cutting memory usage by 50%. • Managed end-to-end data ingestion using Snowflake (warehouse), Informatica (ETL mapping), UC4-Automic, ServiceNow, and Rally streamlining data accessibility across the EDCP project. • Optimized the Technical Analyst SharePoint documentation site, reducing navigation time by 30% through improved structure and updated data governance standards.
• Fine-tuned YOLOv5 on custom PCB image datasets for surface defect detection (cracks, soldering errors, missing components) achieving 91% mAP and reducing manual inspection time by 40%. • Designed modular LangGraph pipelines with multi-key state management and reducer-driven aggregation cutting merge bugs by 30% and achieving 40% lower response latency via Groq API integration. • Built a latency-optimized conversational AI agent using FastAPI, Docker, and Kubernetes, scaling to 50,000+ daily queries with Redis caching reducing average response time by 75% (4.2s to 1.1s). • Engineered a Healthcare Domain Chatbot using LangGraph, fine-tuned LLM (Q-LoRA Phi-3), and Astra DB vector store evaluated with SHAP, ROUGE, BERT Score, and RAGAS for production-grade quality.