Machine Learning Engineer · Data Science at the core

Good | I'm Yashwanth Gowram

A machine learning engineer with 4+ years turning complex data into production-grade ML systems across financial services and FAANG e-commerce. From a Master's in Data Science at Florida State University to fraud-risk scoring and demand forecasting, I build models that survive production, drift, and compliance sign-off.

Yashwanth Gowram
Yashwanth Gowram
Machine Learning Engineer
M.S. Data Science · FSU
Skills

Core skills & libraries that power my work

Career Journey

The experience and education

A timeline of growth from a Master's in Data Science to building production ML systems for fraud, risk, and forecasting.

Machine Learning Engineer, Contract

Capital One
06/2025 – 05/2026
Tallahassee, FL (Remote)

Fraud-risk transaction scoring, data-drift monitoring, and TB-scale PySpark pipelines in a regulated, audited environment.

Master of Science in Data Science

Florida State University
08/2023 – 05/2025
Tallahassee, FL

Focused on machine learning, statistical modeling, and large-scale data systems. CGPA 3.86 / 4.0.

Machine Learning Engineer

Amazon
08/2020 – 07/2023
Bengaluru, Karnataka, India

XGBoost demand forecasting on Spark pipelines and ML platform work across e-commerce catalog systems.

Bachelor of Technology in Electronics & Communication Engineering

Gokaraju Rangaraju Institute of Engineering and Technology
Hyderabad, India
Microsoft Certified · Azure Data Scientist Associate Microsoft Certified · Power BI Data Analyst Associate Google Cloud · Professional Machine Learning Engineer Esri · ArcGIS Pro Basics
Projects

The systems and pipelines I build to ship, monitor, and scale.

From retrieval-augmented assistants to real-time fraud scoring, each project is engineered for performance, reliability, and measurable business impact.

AI Systems · NLP RAG PIPELINE ingest → embed → retrieve → answer Documents PDF · DOCX · TXT Databricks Ingest + Chunking Embeddings Transformers · 768-dim MLflow Eval · Precision FAISS IVFPQ Vector Search LangChain RAG + Source Cite Streamlit Grounded Responses

Document Intelligence Pipeline

Grounded RAG + semantic search over unstructured docs
Core Stack
PythonHF TransformersFAISSLangChainMLflow+2 others
View Details → FULL STACK ⟳

Document Intelligence Pipeline

A low-latency retrieval-augmented pipeline over unstructured documents. Hugging Face transformers produce 768-dimensional embeddings on an optimized FAISS IndexIVFPQ, while LangChain generates grounded, source-cited answers and MLflow tracks retrieval precision across runs.

Impact
  • Retrieval <50ms vector search
  • 768-dim embeddings on IVFPQ
  • Grounded, source-cited answers
System Components
  • Semantic Retrieval Pipeline
  • FAISS IndexIVFPQ Store
  • LangChain RAG + Grounding
  • Databricks Ingest + Chunking
  • MLflow Eval Framework
↺ Click to flip back
Fraud · Risk ML Automated Retraining TRAINING + CI/CD Azure ML SMOTE + Cost-Sens XGBoost·sklearn recall 68 → 90% MLflow 30+ runs tracked GH Actions Auto-Retrain deploy INFERENCE + ANALYSTS Raw Logs Feature Eng Velocity + Identity Scoring Model SHAP Signals Streamlit Analyst Risk UI

Real-Time Fraud Detection Pipeline

End-to-end fraud scoring with explainability
Core Stack
PythonXGBoostscikit-learnSHAPAzure ML+3 others
View Details → FULL STACK ⟳

Real-Time Fraud Detection Pipeline

End-to-end fraud scoring that tackles severe class imbalance with SMOTE and cost-sensitive learning. Velocity and identity-clustering features feed an XGBoost model, with SHAP explanations surfaced to analysts through a live risk-scoring interface and GitHub Actions driving retraining.

Impact
  • Minority recall 68 → 90%
  • 30+ experiment iterations tracked
  • Audit-ready + auto-retraining
System Components
  • XGBoost + SMOTE Classifier
  • Velocity & Identity Features
  • SHAP Explanation Layer
  • Azure ML Studio Training
  • GitHub Actions Retraining
↺ Click to flip back
GenAI · Text-to-SQL TEXT-TO-SQL self-correcting validation loop User NL Prompt Text-to-SQL Dynamic Schema Inject FastAPI SELECT name, SUM(rev) FROM orders WHERE … Syntax Validation Filter Valid SQL Re-prompt invalid · 35 → 9% SQLite results Prompt-Eng Iterations · Spider Benchmark 74% execution accuracy · 20+ MLflow runs on Databricks

Intelligent SQL Query Assistant

Self-correcting Text-to-SQL over open LLMs
Core Stack
PythonLangChainHF TransformersFastAPISQLite+2 others
View Details → FULL STACK ⟳

Intelligent SQL Query Assistant

Text-to-SQL over open-source LLMs with dynamic schema injection. A self-correcting validation loop intercepts invalid SQL at inference time and re-prompts the model, with every prompt iteration tracked in MLflow on Databricks and scored on the Spider benchmark.

Impact
  • Execution accuracy 48 → 74%
  • Malformed queries 35 → 9%
  • 20+ MLflow-tracked runs
System Components
  • Dynamic Schema Injection
  • Syntax Validation Loop
  • FastAPI Inference Service
  • SQLite Execution Layer
  • MLflow Run Tracking
↺ Click to flip back
More forecast-svc python · airflow drift-monitor mlflow · alerts rag-eval-kit langchain · faiss sku-classifier sagemaker

More on GitHub

Forecasting services, drift monitors, and eval kits
Core Stack
PythonAirflowMLflowSageMakerand more
Open GitHub → GITHUB ⟳

Let's build something.

Open to ML engineering & data science roles.