Resume

• Generated actionable business insights by collecting, cleaning, and analyzing structured and unstructured datasets using exploratory data analysis (EDA), supporting data-driven decision making.
• Enhanced Generative AI solutions by designing prompt engineering and evaluation workflows, integrating LLM APIs, and developing dashboards and analytical reports for cross-functional stakeholders.

• Conducted AI market research and competitive analysis on 50+ major market players to identify opportunities for product development and feature innovation.
• Created 10+ technical presentations and research reports translating AI concepts into actionable business recommendations.

Python, Ollama, Qwen, LLMs, Multi-Agent Systems, Prompt Engineering, Agent Orchestration, Software Engineering, Pytest

• Automated end-to-end software development by building a fully local multi-agent platform that converts natural language requirements into implementation-ready projects using Analyst, Architect, Planner, Critic, Coder and Fixer agents.
• Improved software quality by designing a consensus-driven orchestration pipeline with dependency-aware planning, context-aware code generation, automated compilation, pytest-based testing, and iterative self-repair.
• Enabled offline AI-assisted development through custom agent orchestration, MD5 response caching, dependency-aware context management, and Ollama-powered local LLM inference for reproducible software generation.

Python, PyTorch, TensorFlow, EfficientNet, 3D U-Net, MRI Processing, Medical AI, BraTS 2020, Mixed Precision

• Improved MRI brain tumor segmentation by developing and evaluating three 3D architectures—Baseline U-Net, EfficientNet U-Net, and Attention-EfficientNet U-Net—on the BraTS 2020 dataset.
• Enhanced volumetric learning by implementing multimodal MRI preprocessing, tumor-aware patch sampling, mixed-precision training, and Dice + Cross-Entropy optimization for large-scale 3D segmentation.
• Increased segmentation performance from 80.57% Dice to 82.56% Dice (75.47% IoU) while conducting comparative analysis of EfficientNet encoders and attention mechanisms.

Python, TensorFlow, Keras, CNN, MFCC, Librosa, Speech AI, Audio Signal Processing, Deep Learning

• Recognized emotions from speech by developing a CNN-BiLSTM model using MFCC feature extraction, achieving 72% classification accuracy on the RAVDESS and TESS datasets.
• Built a robust audio processing pipeline for feature extraction, preprocessing, and model training on high-dimensional speech datasets using TensorFlow and Librosa.
• Validated the research contribution by authoring a 6-page research paper accepted for presentation at the IEEE Intl. Conference on Data, Energy and Communication Networks (DECON).