Ravi Teja Kothuru

Senior Operations Specialist | CITI Bank (CSIPL) | February 2023 - September 2023

• Financial Processing of ACH Transactions: Managed the daily processing of high-volume ACH transactions according to NACHA guidelines, ensuring compliance and timely completion.
• Fraud Detection and Resolution: Identified and returned fraudulent or unauthorized transactions, minimizing financial losses for both clients and the bank. Collaborated with the fraud department to enhance detection methods and prevent future occurrences.
• Client Support and Transaction Management:
• Assisted clients in blocking or removing blocks on transactions from specific merchants based on their requests.
• Initiated funds transfers to other financial institutions upon receipt of a valid Letter of Indemnity, maintaining accurate records and ensuring adherence to regulations.
• Resolved client queries related to transaction histories, held funds, returned items, and other related issues by providing clear and concise information.
• Issue Investigation and Resolution:
• Investigated missing transactions, payment discrepancies, and other problems to address client or creditor concerns effectively.
• Referral of unresolved grievances to designated departments for further investigation when necessary.
• Process Improvement:
  Developed and implemented Python scripts to automate and validate high-volume ACH transactions, reducing manual intervention and improving data accuracy.
  Spearheaded an initiative to streamline ACH transaction processing workflows, resulting in a 35% reduction in manual efforts and improved SLA compliance by 40%. This achievement was accomplished through close collaboration with automation, fraud, and other relevant departments.
  Led the automation of key operational processes, resulting in a 50% reduction in manual efforts and improved SLA compliance by 45%.
  Spearheaded the implementation of an robotic automation process (RPA) solution to automate ACH transaction processing, reducing errors by 40% and enhancing processing times by 60%.
  Developed and maintained KPI dashboards using Tableau and Power BI, enabling real-time monitoring of process performance and facilitating proactive issue resolution.
  Participated in various projects focused on digital transformation, contributing to the successful migration of operations to a robotic process automation (RPA) environment.
• Stakeholder Management and Collaboration:
  Worked closely with various teams (fraud, automation, operations) to plan and coordinate changes that resolved client transaction problems efficiently.
  Served as a key point of contact for clients, ensuring their needs were met and maintaining strong relationships.
  Trained and mentored new team members in process automation, data analysis, and operational excellence.

Operations Specialist | CITI Bank | March 2021 - January 2023

• Assisted in daily ACH transaction processing tasks, ensuring timely completion and high-quality standards.
• Contributed to process improvement initiatives focused on ACH transactions, resulting in a 30% reduction in task processing times.
• Worked closely with the data analysis team to extract insights from operational data, supporting strategic decision-making related to ACH processing.
• Assisted in regulatory compliance efforts by ensuring adherence to NACHA Operating Rules and other relevant guidelines.

Medical Image Classification

Objective: Developed a CNN-based image classification model to assist medical professionals in identifying pathologies within X-ray images.

Approach:

Explored and preprocessed datasets containing over 50,000 labeled chest X-ray images from various sources (e.g., RSNA pneumonia detection challenge, MIMIC-CXR).

Applied data augmentation techniques (random rotation, zoom, shift, flip, crop) to create a more robust model.

Trained multiple CNN architectures (VGG16, ResNet50, DenseNet, InceptionV3) using transfer learning and fine-tuning approaches.

Evaluated models' performance using metrics such as accuracy, precision, recall, F1-score, and AUC-ROC on validation sets.

Integrated the best-performing model with a medical imaging viewer for real-time analysis and diagnosis.

Outcome: Achieved high accuracy (95%+) in detecting various pathologies, enabling medical professionals to make more informed decisions based on the classifications provided.

Predictive Maintenance System

Objective: Built a predictive maintenance system using time-series analysis and machine learning algorithms for industrial machinery to minimize downtime.

Approach:

Collected data from IoT sensors installed on 50+ pieces of machinery, capturing metrics like temperature, pressure, vibration, and noise levels.

Performed exploratory data analysis (EDA) to identify trends, patterns, and outliers in the collected data.

Implemented feature engineering techniques such as lag creation, differencing, and exponential moving averages to capture temporal dependencies.

Trained various time-series models (ARIMA, SARIMA, LSTM, GRU, Prophet) using historical data to predict equipment failures.

Evaluated model performance using metrics like RMSE, MAE, R-squared, and mean absolute percentage error (MAPE).

Implemented an alert system that triggers maintenance activities based on predicted failure times, reducing unexpected downtime by 40%.

Outcome: Successfully implemented a predictive maintenance system that reduced downtime and maintenance costs by enabling preventative scheduling.

Real-Time Fraud Detection System

Objective: Implemented a real-time fraud detection system for online transactions using historical data and anomaly detection in user behavior.

Approach:

Analyzed transactional data containing features such as transaction amount, location, device type, time of day, and user history.

Applied dimensionality reduction techniques (PCA, t-SNE) to visualize and understand the dataset better.

Trained unsupervised learning models (One-Class SVM, Isolation Forest, Local Outlier Factor) to detect anomalous transactions.

Evaluated model performance using metrics like precision, recall, F1-score, and ROC curve for imbalanced datasets.

Implemented a real-time transaction monitoring system that flags suspicious activities for further investigation.

Outcome: Improved fraud detection by 35% compared to the previous rule-based system, reducing false positives and negatives.

Chatbot Development

Objective: Developed a conversational AI system using Python and NLP techniques to automate customer support for an e-commerce platform.

Approach:

Explored intents, entities, and dialogue acts from customer queries to create a structured dataset for training NLP models.

Trained sequence-to-sequence (Seq2Seq) models with attention mechanisms using LSTM and GRU architectures to generate human-like responses.

Implemented word embeddings (Word2Vec, GloVe, BERT) to capture semantic meaning and improve response generation.

Evaluated model performance using metrics such as perplexity, BLEU score, ROUGE-L score, and human evaluation for coherence and relevance.

Integrated the chatbot with existing customer support platforms for 24/7 assistance and seamless handoff to human agents when required.

Implemented context-aware conversation management to maintain a consistent dialogue flow across multiple user interactions.

Outcome: Successfully deployed the chatbot, reducing customer wait times by 60% and lowering the workload on human support agents by handling simple queries efficiently. The chatbot maintained high satisfaction scores (85%+) based on user feedback.

5.E-commerce Recommendation System

Objective: Built a recommendation engine using collaborative filtering and content-based methods to provide personalized product suggestions for e-commerce users.

Approach:

Analyzed user behavior data, including browsing history, click-throughs, purchases, and ratings, along with item metadata (e.g., category, tags, description).

Applied matrix factorization techniques like Singular Value Decomposition (SVD) and Non-negative Matrix Factorization (NMF) for collaborative filtering.

Trained content-based models using word embeddings (Word2Vec, GloVe), TF-IDF, and BERT to capture item similarity based on metadata.

Combined collaborative and content-based approaches using hybrid models like Weighted Regularized Matrix Factorization (WRMF) and Model Fusion for improved recommendations.

Evaluated model performance using metrics such as precision at K, recall at K, mean average precision (MAP), and normalized discounted cumulative gain (NDCG).

Implemented real-time updates to recommendation lists based on new user interactions or added products.

Outcome: Achieved a 25% increase in customer engagement with personalized product recommendations, leading to a 10% rise in sales conversions. The recommendation system also improved the site's SEO by driving more targeted organic traffic.