TAKSHAK PATIL

I am a highly motivated AI/ML Engineer. I have a solid foundation in computer science. My expertise includes ethical bias detection, deep learning, and advanced Python programming. I have shown real success in hands-on work. This involves designing fairness-aware pipelines and developing multi-modal fusion architectures. I also deploy bias mitigation strategies for medical AI applications. I am skilled at using top frameworks like TensorFlow, PyTorch, and Scikit-learn. My practical projects cover healthcare diagnostics, generative gaming platforms, and real-time gesture recognition. People recognize my excellence in technical troubleshooting and collaborative research. I focus on data-driven problem-solving too. At the same time, I pursue certifications and offshore opportunities. This helps me expand my skills and gain a broader global perspective.

The major project focused on the Ethical AI Bias Simulator using the BRSET Diabetic Retinopathy Dataset.

Major Project Achievements.

I put together a fairness-aware deep learning pipeline based on an EfficientNet-B0 backbone. That backbone had over 4.7 million parameters. It hit a test AUC of 0.8868 along with a test accuracy of 89.74 percent. The highest validation accuracy reached 95.58 percent.

I came up with a multi-modal fusion architecture that pulled in fundus image data and clinical tabular data. This setup helped optimize detection for diabetic retinopathy across a big medical dataset. That dataset included 16,266 samples.

I rolled out several bias mitigation strategies like threshold optimization, sample reweighting, and adversarial fairness constraints. These steps cut down the demographic parity difference to pretty low levels. For instance, the age group DP difference dropped to 0.043.

I carried out full fairness evaluations for subgroups based on age, gender, and equipment. All this provided useful insights into managing bias when deploying AI in clinical settings.

I showed strong technical skills in building a scalable and production-ready design. The training ran efficiently on GPU with CUDA acceleration. Evaluations used solid cross-validation methods. I also created thorough model exploitability tools suited for healthcare use.

In the end, I delivered a solid AI solution that others could replicate. It included clear recommendations for monitoring fairness in real clinical practice. There were also notes on validating across different datasets or in clinical trials.

• successfully completed the CITI Program "Human Research" course (Basic, Data or Specimens Only) with a score of 95, under the institutional affiliation of Massachusetts Institute of Technology Affiliates, on April 4, 2025. This is verified by your completion certificate and report, demonstrating compliance with research ethics and human subject .
  
• This achievement establishes connection with the CITI Program, and affiliation is listed as Massachusetts Institute of Technology Affiliates in both the detailed report and the official certificate.

• Microsoft Azure Fundamentals: Describe cloud concepts
• Cisco Certified Network Associate CCNA v7
• Machine Learning with Python IBM
• Database Foundations Oracle Academy
• Ideathon 2nd Position - Think-On
• CITI - Data or Specimens Only Research Massachusetts Institute of Technology Affiliates

AI Ethical Bias Simulator on BRSET Diabetic Retinopathy Dataset , Major project Jan 2025 — May 2025, Technologies: Python, PyTorch, TensorFlow, Scikit-learn, Fairlearn, EfficientNet, OpenCV, Engineered a fairness-aware deep learning pipeline with EfficientNet-B0 backbone (4.7M parameters) achieving 94.93% accuracy while reducing demographic parity difference to 0.043 across protected attributes in medical AI diagnosis, Developed multi-modal fusion architecture combining fundus image analysis with clinical tabular data among 16,266 patients, achieving 89.74% test accuracy and 88.68% AUC with equitable performance across demographic subgroups, Implemented comprehensive bias detection framework using statistical hypothesis testing (Chi-square, ANOVA) identifying significant demographic disparities across age (Chi-square: 29.84, p95.58% validation accuracy and 0.14% AUC improvement while maintaining fairness evaluation metrics, Deployed GPU-accelerated training with equipment-specific augmentation strategies addressing domain generalization challenges across Canon CR and Nikon imaging systems, reducing bias from 12.5% to 0.5% across protected subpopulations, 

AI-Driven Generative Platformer, Minor Project Jan 2024 — May 2024, Develop an AI-driven system that utilizes the Stable Diffusion model for image generation and a large language model (LLM) for dynamic character creation and interaction. This project aims to enhance the game development process by providing players with unique, personalized in-game characters and engaging NPC interactions, ultimately improving player engagement and immersion., 

Crew AI Property Research Sep 2022, Directed the creation of an AI-driven property research crew using Crew AI and Ollama's NLP capabilities.,

Brain Tumor Prediction Mar 2023, Developed ML models with MobileNet and VGG16, applied diverse preprocessing and hyper-parameter tuning techniques, achieving 97% accuracy., 

Real-time Gesture Recognition System Sep 2023, Developed a real-time action detection system using OpenCV and MediaPipe. Holds promise for enhancing communication accessibility for the hearing-impaired people, potentially revolutionizing inclusive technology solutions., 

Stock prediction using time series analysis Dec 2023, Utilized time series analysis to model historical stock price data, identifying patterns and trends. Developed predictive models to forecast future stock prices.

• Verbal Communication
• Leadership and Managing
• Time Management
• Problem-Solving