Sai Chandu
Machavarapu.
Engineering AI systems that actually ship.
Results-driven AI & ML Engineer with 4+ years of hands-on experience designing, building, and deploying production-grade machine learning systems across banking, healthcare, and manufacturing domains. Skilled across the full ML lifecycle including data ingestion, ETL, feature engineering, model training, experiment tracking, deployment, monitoring, and retraining. Specialized in Generative AI, LLM engineering, RAG pipelines, LangChain, LlamaIndex, vector databases, cloud AI platforms, and MLOps automation.
- 4+
- Years in production ML
- 3
- Domains: banking, health, mfg
- 11
- Industry certifications
Where I've worked.
Banking, healthcare, and manufacturing — building reliable models, RAG systems, and MLOps pipelines.
- Nov 2025 – Present
Comerica
Dallas, Texas, USAAI / Machine Learning Engineer- Designed scalable end-to-end AI/ML pipelines using Python, SQL, Apache Spark, and Apache Airflow for enterprise banking and financial analytics.
- Developed ML and deep learning models using TensorFlow, PyTorch, Scikit-learn, XGBoost, and LightGBM for fraud detection, credit risk scoring, anomaly detection, and forecasting.
- Built Generative AI and LLM solutions using OpenAI API, Azure OpenAI, Hugging Face, LangChain, RAG, Prompt Engineering, LoRA/PEFT, FAISS, Pinecone, and ChromaDB.
- Implemented NLP pipelines for text classification, sentiment analysis, NER, semantic search, topic modeling, and automated document processing.
- Built MLOps frameworks using MLflow, Kubeflow, Docker, Kubernetes, CI/CD, Jenkins, and DVC.
- Developed REST APIs and AI microservices using FastAPI and Flask integrated with AWS SageMaker, Lambda, S3, and GCP Vertex AI.
- Implemented model monitoring, drift detection, explainability, SHAP, LIME, Prometheus, and Grafana.
- Nov 2024 – Oct 2025
Memorial Hermann Health System
Houston, Texas, USAMachine Learning Engineer- Designed ML and AI solutions for healthcare analytics using Python, SQL, TensorFlow, PyTorch, and Scikit-learn.
- Built LLM applications using OpenAI, Hugging Face, LangChain, RAG, FAISS, and Pinecone for clinical document search, summarization, and conversational AI.
- Developed deep learning models using CNNs, RNNs, LSTMs, Transformers, XGBoost, and LightGBM for patient risk prediction and outcome forecasting.
- Designed LangChain agents for document authoring, semantic search, and clinical recommendation workflows.
- Deployed healthcare AI workloads using MLflow, Kubeflow, Docker, Kubernetes, FastAPI, Flask, GCP Vertex AI, and Azure ML.
- Built ETL and real-time processing frameworks using Apache Spark, Kafka, Airflow, PostgreSQL, and MongoDB.
- Apr 2022 – Jul 2024
Equitas Small Finance Bank
Chennai, IndiaData Scientist- Developed predictive models using Python, Scikit-learn, TensorFlow, XGBoost, and LightGBM for credit risk, fraud detection, customer segmentation, and forecasting.
- Built ETL pipelines using Apache Spark, Pandas, PostgreSQL, and MongoDB.
- Implemented NLP solutions using spaCy, NLTK, Transformers, and Hugging Face for feedback analysis, document classification, and sentiment analysis.
- Deployed model services using FastAPI, Flask, Docker, Kubernetes, MLflow, and AWS SageMaker.
- Built Tableau and Power BI dashboards for business intelligence and executive reporting.
- Mar 2020 – Mar 2022
MRF Tyres
Chennai, IndiaData Scientist- Developed predictive analytics and machine learning models using Python, Scikit-learn, TensorFlow, PySpark, and Spark MLlib for demand forecasting, quality defect prediction, anomaly detection, and operational optimization across tyre manufacturing.
- Built scalable big data pipelines using Apache Spark, Kafka, Azure Databricks, Hive, and Hadoop to ingest and process large-scale manufacturing sensor and operational datasets in near-real-time.
- Applied feature engineering, PCA, cross-validation, and hyperparameter tuning with ROC-AUC and precision-recall evaluation; performed customer segmentation and time-series forecasting to optimize inventory and supply chain planning.
- Delivered interactive dashboards using Tableau, Power BI, Matplotlib, and Seaborn for production KPIs, model performance tracking, and cross-functional stakeholder reporting.
Things I've built.
Real-time credit-card fraud scoring with a multi-agent LLM investigator and per-transaction explainability. Calibrated XGBoost + LightGBM ensemble (isotonic, 55/45 blend) at 0.996 AUC-ROC, 0.963 F1, and 95.4% fraud recall at a 0.4% false-positive rate. LangGraph multi-agent investigator with weighted voting, RAG-grounded policy citations, and deterministic fallback. Served sub-5ms scoring via FastAPI (4.3ms p99, 28× under SLA) with per-request SHAP explanations, plus a Streamlit dashboard and MLflow registry.
Retrieval-augmented Q&A over clinical notes and medical guidelines with hallucination detection for clinical safety. Hybrid dense + sparse retriever (ChromaDB + BM25 with ICD-10/CPT-aware tokenization, fused via RRF) reaching 0.92 recall@5 (+15.7% over dense-only). Semantic chunking cut the hallucination rate 82% (21.2% → 3.8%). Multi-layer hallucination detection (medical-term overlap, ROUGE-1, RAGAS-style context precision) with citation grounding over a 30-doc corpus at 36ms p95. End-to-end ingestion, retrieval, generation, FastAPI service, Streamlit UI, and 23 pytest tests.
End-to-end MLOps platform for industrial sensor forecasting, anomaly detection, drift monitoring, and automated retraining. Trained an attention LSTM forecaster (2.31% MAPE vs 4.5% target), an Isolation Forest detector (0.884 F1), and an XGBoost anomaly classifier (0.963 macro-F1) across 8 sensors and 3 years of hourly data. Engineered 270+ features with strict train-only scaling. PSI-based drift monitoring caught 3/3 injected drifts; APScheduler retraining promotes models only on ≥2% improvement. Served via FastAPI (4.3ms p95) with a 5-page Streamlit dashboard and MLflow registry.
Two-stage retrieval + ranking recommender with a feature store and a built-in A/B testing framework. ALS collaborative filtering + TF-IDF content candidates feeding a LightGBM LambdaMART ranker reaching 0.829 NDCG@10 (+23.5% over CF). 3-arm A/B test (consistent-hash, chi-square + Welch's t-test) drove +33.8% CTR (p ≈ 1e-27) and +39% revenue/user. Redis feature store with L1/L2 caching sustained 251 RPS at 84.5% hit rate, 0.095ms p95. Leave-one-out evaluation, cold-start fallbacks, async SQLite event pipeline, MLflow tracking, and a 5-page Streamlit analytics dashboard.
The toolkit.
Hands-on across the full ML lifecycle — from data and modeling to deployment, monitoring, and Generative AI.
01 — Programming
02 — AI / ML
03 — Generative AI & LLMs
04 — NLP & Computer Vision
05 — Frameworks & Libraries
06 — MLOps & Deployment
07 — Cloud
08 — Data Engineering
09 — Databases
10 — DevOps & Tools
11 — Monitoring & Visualization
Academic background.
M.S. Computer Science & Information Systems
Verified across AI, cloud & data.
Let's build intelligent, production-ready AI systems.
Available for AI/ML, GenAI, MLOps, and Data Science roles. Reach out about AI/ML, GenAI, RAG, or MLOps roles.