Suraj Kumar
LLM & Agents
4+ years exp
4+ years ML exp
Management: Under a year
10+ projects
C++
Python
Data Pipeline & Feature Engineering: 4/10
Experimentation & Evaluation: 4/10
MLOps & Deployment: 4/10
Active 3 days ago
Download CVCV
Overview
Technical skills
Projects
Timeline
Roles
Overview
Aspiring AI Engineer with hands-on experience in Generative AI, Large Language Models (LLMs), Machine Learning, and MLOps. Skilled in building RAG applications, AI agents, and scalable ML deployment pipelines using Python, LangChain, Docker, Kubernetes, and AWS. Seeking to leverage technical expertise to develop innovative AI-driven solutions and contribute to impactful business outcomes.
Phone
Technical skills
Languages
2
C++
Python
Databases
4
MongoDB
PostgreSQL
Chroma
Amazon DocumentDB
AI/ML
7
LLM
Machine Learning
MLFlow
MLlib
RAG
LLMOps
Semantic Search
Other
16
CI/CD
AWS
AIOps
Boto3
NumPy
Pandas
Scikit-learn
XGBoost
Kubernetes
LangChain
GitHub
Jupyter Notebook
DVC
Docker
Docker Compose
AI Agents
Projects
Machine Learning & MLOps Projects – End-to-End ML Solutions
Developed multiple machine learning projects focused on solving practical prediction and data analysis problems using Python and machine learning techniques. Implemented data preprocessing, feature engineering, model training, evaluation, and prediction workflows using standard ML libraries. Explored MLOps practices to improve workflow reproducibility, model deployment, experiment tracking, and maintainability. Organized project code and documentation in GitHub to support reproducibility and continuous improvement.
GitHub repository: https://github.com/Suraj Kumar09/MLOPs_Machine_Learning_Projects
TripMate AI – Multi-Agent Travel Planner with LangGraph and MCP
Developed an end-to-end AI-powered travel planning application using a multi-agent architecture orchestrated with LangGraph. Integrated specialized AI agents for travel planning, budget management, and information gathering, with MCP-based tools for flight and hotel searches and weather information. Implemented a supervisor agent for dynamic task routing, input guardrails for safer interactions, and Human-in-the-Loop (HITL) approval with interrupt-and-resume workflows. Built a FastAPI web application with PostgreSQL-based conversation state persistence to support structured, stateful travel planning.
GitHub repository: https://github.com/Suraj-Kumar09/TripMate-AI-A-Multi-Agent-Travel-Planner-with-LangGraphs
Medical AI Assistant – RAG-Based LLMOps Application
Developed a Retrieval-Augmented Generation (RAG) based Medical AI Assistant to provide context-aware responses using relevant medical documents. Implemented a document processing pipeline involving text extraction, chunking, embeddings, vector database storage, and semantic similarity search to retrieve relevant information. Integrated an LLM with retrieved context to generate grounded responses and reduce unsupported outputs. Applied LLMOps practices to structure the application pipeline, manage prompts, and support containerization and deployment workflows.
GitHub repository: https://github.com/Suraj-Kumar09/Medical-AI-Assistant-LLMOps
Developed an end-to-end Machine Learning Operations (MLOps) pipeline to predict customer churn using Python and machine learning techniques. Implemented data preprocessing, model training, evaluation, and prediction workflows, with a focus on reproducibility and maintainability. Integrated MLflow for experiment tracking, Docker for containerization, and CI/CD automation to streamline model development and deployment. Designed the pipeline to support reliable predictions and simplify the management of the machine learning lifecycle.
GitHub repository: https://github.com/Suraj-Kumar09/Customer_Churn_Prediction_MLOps_Pipeline
Timeline
Guru Gobind Singh Indraprastha University (GGSIP University)
Bachelor's Degree ·
Bachelor of Technology (B.Tech), Artificial Intelligence & Data Science
LLM Training Intern
·
Junior
Ethara.ai
·
Internship
Evaluated Large Language Models (LLMs) to enhance response quality, consistency, and factual accuracy by applying advanced prompt engineering and instruction tuning techniques across multiple evaluation tasks. By utilizing structured evaluation methodologies, I systematically reviewed AI-generated outputs, collaborated with cross-functional AI teams to optimize model performance, reliability, and alignment, and directly contributed to the continuous improvement and robustness of Generative AI systems.
Open Source Contributor
·
Middle
Winter of Code Social
·
Internship
Developed end-to-end AI-powered applications and machine learning workflows using Python, covering everything from data preprocessing and model development to testing and scalable deployment. By adhering to software engineering best practices and Git-based version control, I collaborated closely with development teams to deliver reliable, maintainable, and high-performance AI systems.
AI & Software Developer Intern
·
Junior
UniHox Tech
·
Internship
Collaborated with mentors and distributed global teams on GitHub to develop new features, resolve complex issues, and contribute directly to open-source projects. I consistently improved code quality, system reliability, and maintainability by conducting comprehensive peer code reviews, writing detailed documentation, and implementing rigorous testing practices.
Middle AI/ML Engineer
Confidence: High ML Engineer
MLOps-focused ML engineer at a Middle level with a strong track record building end-to-end classical ML pipelines and experiment tracking. The strongest proven skill is reproducible training and evaluation engineering with MLflow and structured pipeline components, evidenced by exp2_bow_vs_tfidf.py, src/model/model_evaluation.py and us_visa/cloud_storage/aws_storage.py. There is little public evidence of custom neural architectures, large-scale distributed training, GPU optimizations or production-grade CI/CD manifests.
Model Architecture & Training
3/10
How well models are designed and trained
Solid use of classical ML training patterns, hyperparameter search and experiment logging but no custom model architectures or advanced training systems.
Evidence
Capstone-Project/notebooks/exp3_lor_bow_hp.py: GridSearchCV + LogisticRegression training and MLflow logging
Capstone-Project/notebooks/exp2_bow_vs_tfidf.py: train_and_evaluate loop with multiple algorithms and mlflow nested runs
MLOPs_Machine_Learning_Projects/CREDIT_CARD_FRAUD_DETECTION/notebook/CREDIT CARD FRAUD DETECTION.ipynb: large training loop with ensemble models and evaluation
Data Pipeline & Feature Engineering
4/10
How data is prepared for models
Multiple data ingestion and feature engineering components, column transformers and preprocessing pipelines are implemented across projects.
Evidence
Capstone-Project/src/features/feature_engineering.py: apply_bow and load_data functions (feature engineering module)
MLOPs_Machine_Learning_Projects/Us_Visa_Prediction/src/components/data_ingestion.py: DataIngestion class for ingestion
MLOPs_Machine_Learning_Projects/Consignment-Pricing-using-MLOps-DVC/notebook/2_Feature_Engineering_and_Model_Tra.ipynb: extensive feature engineering and preprocessing pipelines
Experimentation & Evaluation
4/10
How results are measured and tested
Clear experiment tracking and evaluation practices using MLflow and DagsHub, plus scripts to save metrics and artifacts; notebooks show nested runs and metric logging.
Evidence
Capstone-Project/notebooks/exp1.ipynb: mlflow + dagshub initialization and run logging
Capstone-Project/src/model/model_evaluation.py: evaluate_model, save_metrics, mlflow logging of metrics and model artifacts
Capstone-Project/notebooks/exp2_bow_vs_tfidf.py: nested mlflow runs and mlflow.sklearn.log_model calls
MLOps & Deployment
4/10
How models are shipped to production
MLOps deployment primitives and connectors are present including S3 clients, model pusher/register modules and Flask serving scaffolding, but full CI/CD manifests and production hardening are not consistently present.
Evidence
MLOPs_Machine_Learning_Projects/Us_Visa_Prediction/us_visa/cloud_storage/aws_storage.py: SimpleStorageService S3 helper
Capstone-Project/src/connections/s3_connection.py: s3_operations.fetch_file_from_s3
Capstone-Project/src/model/register_model.py: register_model and MLflow/dagshub integration
Computational Efficiency
2/10
How efficiently computing resources are used
Some practical parallelization and n_jobs usage is present but there is no evidence of GPU optimization, quantization, memory profiling or large-scale performance engineering.
Evidence
MLOPs_Machine_Learning_Projects/CREDIT_CARD_FRAUD_DETECTION/notebook/CREDIT CARD FRAUD DETECTION.ipynb: models created with n_jobs=-1 and parallelized loops
MLOPs_Machine_Learning_Projects/Consignment-Pricing-using-MLOps-DVC/notebook/2_Feature_Engineering_and_Model_Tra.ipynb: use of n_jobs in model training
Research Depth & Innovation
1/10
Depth of research and new ideas
No evidence of novel research, custom layers, paper reproductions or advanced algorithmic contributions; work focuses on applied classical ML and MLOps patterns.
Evidence
Multiple notebooks and scripts showing standard scikit-learn, XGBoost and CatBoost usage rather than custom model implementations (e.g. exp2_bow_vs_tfidf.py, exp3_lor_bow_hp.py)
Expertise
MLOps & Model Lifecycle· Middle
Industries
Financial Services· Middle
Government· Middle
Transportation & Logistics· Middle
Technologies
LangChain
Docker Compose
MLFlow· since 2025 · 5+ projects
AI Agents
AWS
Docker
Kubernetes
GitHub
CI/CD· mentioned only
CI/CD· mentioned only
Docker· mentioned only
DVC· mentioned only
Grafana· mentioned only
minikube· mentioned only
Observability· mentioned only
Prometheus· mentioned only
Recommendations
- Develop and maintain end-to-end ML pipelines that use MLflow for experiments, S3 for model artifact storage and automated model registering/promotion flows.
- Build production model-serving endpoints with FastAPI/Flask and automated deployment pipelines (CI/CD) tied to S3/ECR and Kubernetes, using existing model_pusher and aws_storage modules as starting points.
- Focus on strengthening test coverage and deployment hardening by adding unit and integration tests for key pipeline components and safe secrets handling for tracking URIs.
- If moving toward more advanced ML work, add reproducible distributed training and efficiency engineering (GPU profiling, quantization or batch inference optimizations) for large models.
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories:
Middle Data Scientist
Confidence: Medium ML Practitioner
MLOps-focused ML practitioner (mid-level) with practical experience building end-to-end experiment pipelines and production-flavored model tooling. The strongest proven skill is designing and running ML experimentation and model evaluation pipelines using MLflow/Dagshub integrations and well-structured notebooks, evidenced by exp1/exp2/exp3 notebooks and src/model/model_evaluation.py. Limited evidence exists for hardened production engineering details such as secrets management, robust CI/CD manifests, Spark/big-data tooling, or systematic economic cost analyses of model decisions.
Statistical Rigor
5/10
Correct use of statistics
Shows applied evaluation practice (CV, GridSearch, stratified splits, custom scoring and PR/ROC analysis) but lacks formal uncertainty quantification, multiple-comparison control, causal analysis or rigorous statistical tests beyond standard ML metrics.
Evidence
Capstone-Project/notebooks/exp3_lor_bow_hp.py: GridSearchCV(LogisticRegression(), param_grid, cv=5, scoring='f1')
MLOPs_Machine_Learning_Projects/CREDIT_CARD_FRAUD_DETECTION/notebook/CREDIT CARD FRAUD DETECTION.ipynb: stratified train_test_split and PR/ROC curve evaluation
Capstone-Project/notebooks/exp2_bow_vs_tfidf.py: nested MLflow runs logging metrics for multiple vectorizers and algorithms
Data Wrangling & Cleaning
6/10
Preparing and cleaning data
Consistent, practical data-wrangling and feature engineering across projects with missing-value handling, encoding pipelines, power transforms, and S3/Mongo ingestion code; some pipelines use sensible transformers but there are repeated absolute paths and occasional template code that reduce production-readiness.
Evidence
MLOPs_Machine_Learning_Projects/Us_Visa_Prediction/notebook/2_Feature_Engineering_and_Model_Training.ipynb: feature engineering (company_age), PowerTransformer usage and ColumnTransformer pipeline
Capstone-Project/notebooks/exp1.ipynb and exp2_bow_vs_tfidf.py: text preprocessing utilities (lemmatization, stopword removal, punctuation/url stripping)
Capstone-Project/src/connections/s3_connection.py: class s3_operations with fetch_file_from_s3 (boto3 usage)
Exploratory Analysis & Visualization
6/10
Exploring and visualizing data
Exploratory analysis contains many purposeful visualizations and written insights (chi-sq tests, target imbalance checks, per-feature commentary); notebooks include narrative interpretation after plots rather than just boilerplate charts.
Evidence
MLOPs_Machine_Learning_Projects/Us_Visa_Prediction/notebook/1_EDA_US_visa.ipynb: extensive EDA with written insights and chi2_contingency testing
MLOPs_Machine_Learning_Projects/Consignment-Pricing-using-MLOps-DVC/notebook/2_Feature_Engineering_and_Model_Tra.ipynb: EDA and feature engineering commentary
MLOPs_Machine_Learning_Projects/CREDIT_CARD_FRAUD_DETECTION/notebook/CREDIT CARD FRAUD DETECTION.ipynb: EDA and interpreted results, imbalance analysis and visualizations
Predictive Modeling
6/10
Building models that predict
Modeling shows baseline-first approach, hyperparameter search (GridSearch/RandomizedSearch), resampling for imbalance, cross-validation and model comparison. Some parts use correct CV and nested runs; occasional risky patterns (absolute path usage, resampling ordering unclear in places) limit the rating.
Evidence
Capstone-Project/notebooks/exp3_lor_bow_hp.py: GridSearchCV + nested MLflow logging and best model logging
MLOPs_Machine_Learning_Projects/CREDIT_CARD_FRAUD_DETECTION/notebook/CREDIT CARD FRAUD DETECTION.ipynb: sampling strategies (SMOTE, SMOTEENN), multiple model comparisons and custom weighted scoring
Capstone-Project/notebooks/exp2_bow_vs_tfidf.py: systematic comparison of BoW vs TF-IDF across algorithms with MLflow logging
Business Insight & Impact
3/10
Turning analysis into business value
Some attempts to link analysis to decisions and operational actions (e.g., fraud blocking, model championing), but formal business-metric treatment (cost matrices, FP/FN economic impact, SLA constraints) is limited.
Evidence
MLOPs_Machine_Learning_Projects/CREDIT_CARD_FRAUD_DETECTION/notebook/CREDIT CARD FRAUD DETECTION.ipynb: manual fraud test and explicit statement 'System should block the card'
MLOPs_Machine_Learning_Projects/Vehicle_Insurance/README.md: architecture notes mentioning evaluation threshold checks (score diff > 0.02)
Reproducibility & Notebook Hygiene
4/10
Clean, repeatable analysis
Reproducibility practices present (requirements.txt files, MLflow logging, some tests, setup.py). However notebooks contain absolute local paths, occasional hardcoded tracking URIs, and a number of template/placeholder files which reduce reproducibility and production hygiene.
Evidence
Capstone-Project/requirements.txt and flask_app/requirements.txt: pinned dependency lists
Capstone-Project/notebooks/exp1.ipynb and model_evaluation.py: MLflow/Dagshub integration and logging
Capstone-Project/tests/test_flask_app.py and tests/test_model.py: basic unit tests for flask app and model loading
Expertise
Data Science· Middle
Industries
Financial Services· Middle
Government· Middle
Transportation & Logistics· Middle
Technologies
Python· since 2025 · Middle
MongoDB· since 2023 · 5+ projects
DVC
Jupyter Notebook
XGBoost
Scikit-learn
Pandas
NumPy
Boto3
MLFlow· mentioned only
Recommendations
- Lead implementation of end-to-end model deployment tasks: containerized model serving (Flask/FastAPI) with MLflow/DVC artifacts and S3-backed model storage, using the existing model_pusher and aws_storage modules as starting points.
- Develop production-grade MLOps pipelines: add CI/CD workflows, reproducible infra-as-code, and secrets-safe configuration (avoid hardcoded URIs and local absolute paths present in notebooks).
- Own model evaluation and monitoring workstreams: formalize evaluation thresholds, add calibration/uncertainty reporting, data-drift alerts (Evidently), and automated retrain triggers using the existing model_evaluation and data_validation components.
- Build robust data ingestion and feature-store components for structured data; productionize S3/Mongo ingestion code with backfilling, schema validation, and provenance tracking.
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories:
Middle AI / LLM Engineering (Agents)
Confidence: High Generalist
A generalist ML engineer at a middle level who builds end-to-end ML pipelines and MLOps components. The strongest proven skill is modular MLOps pipeline design and implementation as seen in the customer_churn pipeline and component files such as customer_churn/pipeline/training_pipeline.py and customer_churn/components/*.py. There is limited evidence of production-grade hardening such as automated tests, CI definitions, secrets management and distributed system design.
Architecture & Organization
5/10
How the project is structured
Modular componentized MLOps pipeline with clear dataflow between ingestion, validation, transformation and training, showing deliberate project structure and separation of concerns.
Evidence
Customer_Churn_Prediction_MLOps_Pipeline/customer_churn/pipeline/training_pipeline.py
Customer_Churn_Prediction_MLOps_Pipeline/customer_churn/components/data_transformation.py
Customer_Churn_Prediction_MLOps_Pipeline/customer_churn/components/model_trainer.py
Code Quality & Standards
3/10
How clean the code is
Reasonably readable Python with custom exceptions and logging but limited typing, tests, and some inconsistent patterns and hard-coded paths in notebooks.
Evidence
Customer_Churn_Prediction_MLOps_Pipeline/customer_churn/components/model_trainer.py
Customer_Churn_Prediction_MLOps_Pipeline/customer_churn/components/data_transformation.py
Customer_Churn_Prediction_MLOps_Pipeline/customer_churn/exception/__init__.py
Engineering Culture
3/10
Teamwork and good practices
Signs of engineering culture such as logging, configuration manager and packaging, but missing contribution files, test harness, and formal CI/infrastructure code in the analyzed human files.
Evidence
Customer_Churn_Prediction_MLOps_Pipeline/customer_churn/logger/__init__.py
Customer_Churn_Prediction_MLOps_Pipeline/setup.py
Customer_Churn_Prediction_MLOps_Pipeline/customer_churn/pipeline/training_pipeline.py
Technical Complexity
3/10
How hard the problems solved are
Moderate technical complexity: end-to-end ML pipeline, preprocessing pipelines, resampling, model hyperparameter handling and model bundling but no advanced distributed or systems-level complexity.
Evidence
Customer_Churn_Prediction_MLOps_Pipeline/notebook/Feature_Engineering_and_Model_Training.ipynb
Customer_Churn_Prediction_MLOps_Pipeline/customer_churn/components/data_transformation.py
Customer_Churn_Prediction_MLOps_Pipeline/customer_churn/components/model_trainer.py
Security & Risks
2/10
Protection and risk handling
Minimal explicit security engineering; some use of certifi and dotenv, but there are risky patterns (hardcoded local paths, env var manipulation) and no clear secrets management or hardened deployment configuration in human-authored files.
Evidence
Customer_Churn_Prediction_MLOps_Pipeline/customer_churn/configuration/mongo_db_connection.py
Langchain_Models/langchain-output-parsers/stroutputparser1.py
Reliability & Fault Tolerance
3/10
Stability when things fail
Basic reliability features such as custom exception class, try/except blocks and logging are present, but there is little evidence of automated tests, monitoring, retries or robust fault-tolerance.
Evidence
Customer_Churn_Prediction_MLOps_Pipeline/customer_churn/exception/__init__.py
Customer_Churn_Prediction_MLOps_Pipeline/customer_churn/pipeline/training_pipeline.py
Customer_Churn_Prediction_MLOps_Pipeline/customer_churn/components/model_trainer.py
Industries
Artificial Intelligence· Middle
Commerce· Middle
Telecommunications· Middle
Recommendations
- Lead development of end-to-end ML pipelines and model packaging for production model serving using the existing pipeline components.
- Add unit and integration tests, CI/CD workflows and automated model validation to raise reliability and engineering maturity.
- Implement secure secrets management and remove hard-coded local paths; add deployment manifests or infrastructure-as-code for production environments.
- Prototype scalable model serving (containerized inference, autoscaling, monitoring) to move from prototype to production.
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories: