Home

Manikanta K - Sr. AI/ML Engineer

[email protected]

Location: Remote, Remote, USA

Relocation: YES

Visa: GC

Resume file: MANIKANTA K Resume._1779117552621.docx Please check the file(s) for viruses. Files are checked manually and then made available for download.

MANIKANTA K Senior AI/ML Engineer | LLM & RAG Specialist | Generative AI (469) 708-7552 [email protected] PROFESSIONAL SUMMARY I've been working in ML and data science for about 10 years now, and the last couple of years have been mostly focused on LLMs and Generative AI in production. My background started in classical ML churn models, fraud detection, NLP pipelines and I moved into RAG systems and agentic AI as the tooling matured around 2023. Right now at Fifth Third Bank I'm building RAG pipelines, LangGraph-based agentic workflows, and LLM evaluation frameworks that run at enterprise scale. Before that I was at Tailored Brands doing a mix of traditional ML and early LLM integration from mid-2023. I'm comfortable across the full stack from feature engineering and model training to API deployment, monitoring, and drift detection. I've worked in regulated environments, mentored junior engineers, and know how to explain model outputs to people who don't write code. CORE COMPETENCIES LLM Integration (GPT-4o, LLaMA 3, Mistral) | RAG Pipelines | Graph RAG | Agentic AI | LangChain | LangGraph (2024+) | LangSmith | LlamaIndex | AutoGen | Azure OpenAI | AWS Bedrock | Prompt Engineering | Chain-of-Thought Prompting | Vector Databases (FAISS, Pinecone, Weaviate) | Hybrid Search | RAGAS Evaluation | Hallucination Reduction | NLP & Transformers | BERT | Hugging Face | MLOps | LLMOps | CI/CD for AI | Model Monitoring | Drift Detection | SHAP | LIME | Responsible AI | GDPR & CCPA Compliance TECHNICAL SKILLS LLMs & Gen AI GPT-4 / GPT-4o, Azure OpenAI API, OpenAI API, AWS Bedrock (late 2023+), LLaMA 3, Mistral, Claude API, Prompt Engineering, Chain-of-Thought, Few-shot Prompting, Structured Output (JSON mode), Context Engineering, Token Optimization RAG & Vector Search RAG Pipelines, Graph RAG, FAISS, Pinecone, Weaviate, Qdrant, Chroma, Hybrid Search (BM25 + Dense Vector), Re-ranking (Cohere, CrossEncoder), Sentence-Transformers, RAGAS, LLM-as-Judge, Contextual Compression, Chunking Strategies Agentic AI LangChain, LangGraph (2024+), LlamaIndex, AutoGen, Agentic Workflows, ReAct Pattern, Multi-agent Systems, Tool Orchestration, Function Calling, LangSmith NLP & Transformers BERT, RoBERTa, Hugging Face Transformers, Sentence-Transformers, BERTopic, spaCy, NLTK, Sentiment Analysis, Text Classification, NER, Zero-shot Classification ML / DL Scikit-learn, TensorFlow, PyTorch, XGBoost, LightGBM, CatBoost, Isolation Forest, Autoencoders, LSTM, ARIMA, Prophet, Optuna, SMOTE / ADASYN Explainability SHAP, LIME, Permutation Importance, Model Explainability (XAI), Responsible AI, Model Cards, Bias Reviews, Audit Trails MLOps / LLMOps MLflow, Kubeflow, Apache Airflow, Docker, Kubernetes, Jenkins, GitHub Actions, Weights & Biases, Evidently AI, DVC, Canary Releases, Shadow Deployment, Model Registry, Feature Stores Cloud Azure (ML Studio, Databricks, OpenAI, AI Foundry, Data Factory, AKS), AWS (SageMaker, Bedrock, EC2, Lambda, S3, Glue, EKS), GCP (BigQuery, Vertex AI) Observability LangSmith, RAGAS, Evidently AI, Grafana, Prometheus, SHAP, CloudWatch, Drift Detection, Hallucination Reduction APIs & Services FastAPI, Flask, REST APIs, GraphQL, gRPC, Streaming APIs, Microservices, Real-time & Batch Inference Big Data Apache Spark, PySpark, Kafka, Delta Lake, Apache Iceberg, AWS Glue, Databricks, dbt, Hadoop Programming Python (Expert), SQL, R, Scala, PySpark, TypeScript, Bash Databases MongoDB, PostgreSQL, MySQL, Cassandra, Neo4j, Snowflake, Redshift, Redis, Elasticsearch, Pinecone Visualization Tableau, Power BI, Streamlit, Plotly Dash, Matplotlib, Seaborn WORK EXPERIENCE Senior AI/ML Engineer Jul 2025 Present Fifth Third Bank Cincinnati, OH The main system I own is a production RAG pipeline using Azure OpenAI (GPT-4o) and FAISS with hybrid BM25 + dense retrieval and Cohere re-ranking. It's running across 1M+ daily customer interactions. Support resolution time dropped 35%. The chunking strategy took a few iterations we landed on parent-child chunking after testing fixed-size and sentence-window approaches. On top of the RAG layer I built Graph RAG using Neo4j knowledge graphs and LangChain to ground responses in structured facts. Hallucination rates dropped 23% we measured this using RAGAS faithfulness scores with LLM-as-Judge. Before this the team had no reliable way to know if RAG was actually working. Designed agentic workflows in LangGraph (from early 2024) with ReAct reasoning across internal banking systems. Added human-in-the-loop checkpoints at critical decision nodes. Removed about 60% of the manual steps that analysts were doing daily. Prompt engineering chain-of-thought, few-shot examples, structured JSON output. Response accuracy improved around 28% and token costs came down about 25% at scale. At millions of LLM calls per day that adds up. Set up the LLMOps stack LangSmith for tracing and prompt versioning, RAGAS for quality scoring, Evidently AI for drift, W&B for experiments, MLflow as model registry. Automated retraining cut model degradation incidents by around 80% from baseline. CI/CD for model releases via GitHub Actions and Jenkins automated pytest, Docker builds, canary deployments on Kubernetes. Deployment errors dropped to zero and release cycle shortened by about 38%. Also delivered a churn prediction system using XGBoost with Optuna tuning. Added SHAP dashboards so the business team could understand which factors were driving attrition. Outcome was 23% churn reduction, roughly $5M in annual retention revenue. Stack: Python, Azure OpenAI (GPT-4o), LangChain, LangGraph (2024+), LlamaIndex, AutoGen, FAISS, Pinecone, Neo4j, Graph RAG, Hybrid Search, Cohere Re-ranking, RAGAS, LangSmith, Hugging Face Transformers, BERT, Sentence-Transformers, XGBoost, TensorFlow, PyTorch, Optuna, SHAP, LIME, MLflow, W&B, Evidently AI, Grafana, Apache Spark, Delta Lake, dbt, AWS (SageMaker, S3, EC2, Glue, Lambda), Azure (ML, AKS, Databricks, AI Foundry), Docker, Kubernetes, GitHub Actions, Jenkins, Airflow, FastAPI, PostgreSQL, MongoDB, Neo4j AI/ML Engineer Apr 2023 Jun 2025 Tailored Brands Fremont, CA Fraud detection was one of the first systems I owned here Isolation Forest and Autoencoders in ensemble, deployed as a real-time REST API on Azure ML. Fraudulent transactions dropped 35%, saved about $2M annually. Added SHAP outputs so the risk team had something to show compliance. From mid-2023 we started using Azure OpenAI (GPT-4) seriously. I built the LangChain prompt pipelines and a RAG system using Pinecone for product knowledge retrieval. By the time I left the role it was handling around 65% of tier-1 support without agent involvement. Search relevance improved 41%, zero-result queries dropped 44%. Demand forecasting using ARIMA, LSTM, and Prophet tuned with Optuna, deployed as batch prediction on Azure ML. Planning accuracy up 18%, inventory costs down 15%, stockouts dropped about 19%. Customer segmentation with K-Means and DBSCAN clustering. Enabled targeted campaigns that converted 22% better than untargeted. The marketing team used this directly for their campaign planning. Full MLOps lifecycle Evidently AI for drift, W&B for tracking, DVC for data versioning, GitHub Actions CI/CD, canary rollouts. Model reliability stayed above 99.5% throughout. ETL optimization using PySpark, Azure Databricks, and Delta Lake with dbt. Processing time dropped from about 2 hours to 27 minutes end to end. Downstream ML models got fresher features. Dynamic pricing engine using reinforcement learning and Bayesian optimization via FastAPI. E-commerce revenue up 10%, roughly $3M annually. Mentored five data scientists on MLOps and Azure ML deployment patterns. Stack: Python, SQL, Azure OpenAI (GPT-4, mid-2023+), LangChain, Pinecone, FAISS, Azure ML, Azure Databricks, Delta Lake, Apache Iceberg, dbt, PySpark, Kafka, MLflow, W&B, Evidently AI, DVC, GitHub Actions, Airflow, FastAPI, Docker, Kubernetes, XGBoost, LightGBM, TensorFlow, PyTorch, Optuna, Isolation Forest, Autoencoders, ARIMA, LSTM, Prophet, SHAP, Hugging Face, MongoDB, MySQL, GCP BigQuery, Streamlit Data Scientist Nov 2020 Mar 2023 Charter Communications Maryland Heights, MO Churn prediction for telecom subscribers gradient boosting at 85% accuracy, SHAP-driven feature analysis, automated retraining on SageMaker, Evidently AI drift monitoring. About $2M in subscription revenue retained annually. NLP ticket classification using BERT and RoBERTa on Hugging Face auto-categorized 50K+ monthly support tickets. Response times improved 20%, CSAT up 15 points. The support ops team noticed it within the first month. Recommendation system using collaborative filtering and learned embeddings on SageMaker average order value up 12%, $1.5M annually. Predictive maintenance on IoT sensor data using Isolation Forest. Grafana and Prometheus dashboards for real-time anomaly visibility. Equipment downtime down 19%, saved close to $800K. ML pipeline automation Scikit-learn Pipelines, Optuna, MLflow. Containerized serving via Docker on AWS with Jenkins CI/CD. Cut experiment-to-deployment time by about 30%. Stack: Python, SQL, R, Scikit-learn, XGBoost, LightGBM, TensorFlow, Keras, PyTorch, Optuna, BERT, RoBERTa, Hugging Face, spaCy, SHAP, LIME, MLflow, Evidently AI, Grafana, Prometheus, Apache Spark, Hadoop, AWS (S3, EC2, SageMaker), Azure ML, Docker, Jenkins, Airflow, PostgreSQL, MySQL, Power BI, Tableau Data Analyst Jan 2019 Oct 2020 Change Healthcare Nashville, TN Analyzed 2M+ consumer records in Python and SQL. The patterns we found fed into strategy changes that contributed to about 12% revenue growth across 5 business units. Built sales forecasting models at 88% accuracy, automated recurring reporting in Python saving 10+ hours a week, and designed Tableau and Power BI dashboards tracking 25+ KPIs. Implemented data governance policies and quality audits compliance at 100%, data quality incidents down 40%. Built customer segmentation models using clustering enabled targeted campaigns that converted 18% better than untargeted sends. Stack: Python, R, SQL, Tableau, Power BI, Excel, VBA, Scikit-learn, Hadoop, MySQL, PostgreSQL, AWS (S3, EC2), Azure ML Studio, Matplotlib, Seaborn, Plotly Junior Data Analyst Jun 2015 Aug 2018 RedPine Solutions Hyderabad, India SQL queries and data models for BI reporting across 10+ business units. Reporting accuracy improved 30% after cleaning up data pipeline issues that had been accumulating for a while. Automated data processing using Python and VBA cut manual prep by about 8 hours a week and freed the team up for actual analysis work. Contributed to a customer segmentation project. Targeted campaigns from that work saw about 20% higher engagement than untargeted sends. Stack: Python, R, SQL, Tableau, Excel, VBA, MySQL, PostgreSQL, Git, Matplotlib, Seaborn, AWS EDUCATION Bachelor of Technology, Computer Science 2010 2014 CMR College of Engineering & Technology Keywords: continuous integration continuous deployment artificial intelligence machine learning business intelligence sthree rlang information technology trade national California Missouri Ohio Tennessee