Machine Learning Engineer & Data Scientist
Jhanavi Putcha
Building AI-driven systems for computer vision, deep learning, and production ML
I design and ship ML systems that turn data into measurable business outcomes — from high-throughput computer-vision pipelines to scalable model deployments and LLM-powered applications.
About Me
I'm a Machine Learning Engineer with a Master's degree in Artificial Intelligence, specializing in building production-grade ML systems that solve real-world problems.
My expertise spans industrial inspection automation, predictive analytics, and LLM-backed applications. I'm passionate about model reliability, scalable data pipelines, and delivering measurable business impact through intelligent systems.
Core Expertise
Deep Learning
Neural networks for computer vision, NLP, and predictive analytics
Data Engineering
Large-scale datasets and actionable insights
ML Deployment
Production-ready models with AWS and cloud infrastructure
Generative AI & LLMs
Fine-tuning, RAG pipelines, and GenAI applications
Education
M.S., Artificial Intelligence
University at Buffalo
Computer Vision · Deep Learning · Production ML Systems
B.S., Computer Science
Vignana Bharathi Institute of Technology
Algorithms · Machine Learning · AI Foundations
Technical Skills
LLM & Generative AI
Machine Learning & AI
Cloud & Infrastructure
Programming & Data
MLOps & Deployment
Experience
AI/ML Engineer
Triosoft LLC
Constructed transformer-based NLP models for automated extraction of financial insights from 87+ unstructured reports per week, reducing analyst effort by 18% and improving signal-to-noise ratio in investment research pipelines.
87+ reports/week18% analyst effort reductionEngineered reinforcement learning models for portfolio optimization, simulating 24+ market scenarios per asset class and achieving a 6% improvement in risk-adjusted returns.
reinforcement learning6% improvementImplemented PyTorch/TensorFlow deep learning models for credit risk scoring and market anomaly detection, processing >15 million transaction records monthly and achieving 93% coverage of critical high-risk events.
>15M transactions/month93% coverageDeployed generative AI models and LLMs to automatically generate investment summaries from 80+ financial reports weekly, improving turnaround time by 20%.
80+ reports/week20% fasterOrchestrated end-to-end MLOps pipelines on AWS for training, evaluation, deployment, and drift monitoring, supporting 95+ automated retraining cycles per quarter with <5s latency for real-time scoring.
AWS95+ retraining cycles/quarterFine-tuned LLMs using Hugging Face to automate extraction and summarization for 78+ reports weekly, improving insight generation efficiency by 18%.
Hugging Face78+ reports/weekImplemented a RAG pipeline combining vector DB retrieval with generative AI to produce context-aware investment insights from 59+ documents weekly, improving information accuracy by 15%.
RAG15% accuracy improvement
Machine Learning Scientist
LTIMindtree
Architected time-series forecasting models on Spark and Snowflake for predicting vehicle component failures, improving predictive maintenance accuracy by 18% and reducing unplanned downtime by 12% across 25+ production lines.
SparkSnowflake18% accuracyEngineered real-time anomaly detection pipelines using LSTM and XGBoost on sensor and telemetry data, identifying 83 critical anomalies per month and increasing system reliability by 15%.
LSTMXGBoost83 anomalies/monthDesigned ETL pipelines in Python and PySpark to process 20 TB/month of vehicle data with low-latency ingestion (<5s), enabling 95% data availability for model training.
20 TB/month<5s ingestionBuilt interactive Power BI dashboards surfacing 95+ KPIs across fleet health and production efficiency for 17+ stakeholders.
Power BI95+ KPIsOptimized hyperparameter tuning for LightGBM, XGBoost, and LSTM models, improving forecasting precision by 12% and reducing MAE/RMSE by 15%.
LightGBMXGBoost12% precisionDeployed models on AWS SageMaker with CI/CD and automated retraining, supporting 95+ scheduled model updates per quarter with zero downtime.
SageMakerCI/CDEngineered 25+ features per vehicle from telemetry and maintenance data, improving model performance by 18% and covering 95% of critical components.
feature engineering25+ featuresConstructed a real-time telemetry anomaly detection system using Kafka and Transformer-based time-series models on SageMaker, detecting 87+ critical anomalies per month and reducing unplanned maintenance by 17%.
KafkaTransformer
Featured Projects
A selection of projects showcasing my work in ML, computer vision, and AI applications
Music Generator using Genetic Algorithm
Problem
Most music generators depend on large datasets, making it hard to personalize output without retraining.
Solution
Built a melody generator that evolves music using a genetic algorithm and improves based on iterative selection (fitness).
Outcome
Generates evolving melodies with controllable variation and selection-driven improvement.
Comparative Analysis of RL Algorithms (Discrete & Continuous)
Problem
Choosing the right RL algorithm depends heavily on the action space and environment dynamics.
Solution
Implemented and compared PPO, DQN, DDQN, and A2C across discrete and continuous action-space settings.
Outcome
Analysis highlights trade-offs in stability, sample efficiency, and performance by action space.
Crop Prediction using ThingSpeak
Problem
Farmers need data-driven crop recommendations based on real-time environmental conditions.
Solution
Built a crop prediction workflow using ThingSpeak sensor data (temperature, pH, rainfall, humidity) to guide crop selection.
Outcome
Improved crop selection decisions using real-time sensor-driven insights.
Surya Namaskar Trainer
Problem
Beginners struggle to learn proper Surya Namaskara (Sun Salutation) yoga poses without real-time guidance, leading to incorrect postures and potential injury.
Solution
Built an interactive platform that uses computer vision to provide real-time feedback on yoga pose correctness and alignment.
Outcome
Published research paper (DOI: JETIR2404467) demonstrating effective pose correction for yoga practitioners
Warehouse Robot using Q-Learning and SARSA
Problem
Warehouse logistics require efficient automated systems for picking up and delivering parcels between locations.
Solution
Developed a reinforcement learning agent using Q-Learning and SARSA algorithms to autonomously navigate warehouse environments.
Outcome
Agent successfully learns optimal paths for parcel delivery in complex warehouse grid environments
AI Recipe Generator
Problem
Users often have ingredients but lack inspiration or knowledge to create recipes, especially from visual input.
Solution
Created an AI-powered application that generates food recipes from both image and text inputs, featuring an interactive chatbot powered by OpenAI.
Outcome
Delivers personalized recipe recommendations based on available ingredients through natural conversation
Achievements & Recognition
ML Speaker
Delivered an engaging machine learning-focused talk to an audience of 200+ students , sharing insights on AI/ML concepts and career paths. Received a Letter of Appreciation for knowledge sharing and presentation excellence.
ML Competition Winner
Secured 3rd place in a competitive ML challenge by developing a production-grade AI solution, demonstrating strong problem-solving skills and ability to deliver practical, deployable models under pressure.
Institutional Interface Developer
Built a comprehensive platform that streamlined assignment submissions and improved academic workflows for the institution, showcasing full-stack development capabilities and user-centric design thinking.
Get In Touch
I'm always open to discussing new opportunities, AI/ML projects, or just connecting with fellow tech enthusiasts.