I'm passionate about building intelligent systems that bridge the gap between research and real-world applications. With experience spanning computer vision, NLP, forecasting, and distributed data engineering, I enjoy turning complex problems into efficient, scalable solutions.
⚡ Actively seeking full-time roles (2025) in AI, Machine Learning, or Data Science.

• 🛣️ Designed a real-time pothole detection system(97% precision) using CNN-based computer vision, integrated with GPS tagging for precise location mapping of road defects.
• 🚀 Deployed on NVIDIA Jetson Xavier NX for efficient edge inference with low latency(13ms/frame), eliminating the need for cloud processing.
• 📸 Enabled robust detection under varying lighting and road conditions through data augmentation and transfer learning.
• 📍 Supports automated reporting with GPS-coordinates, facilitating proactive road maintenance and infrastructure planning.
• 📄 Patent granted for the system’s methodology, covering model architecture, detection logic, and deployment pipeline.
• 🏅 Demonstrates a tangible example of AI for social good, improving safety and enabling preventive maintenance for road infrastructure. -Patent

Brystol Myers Squibb & Rutgers University (competition)

• 🧠 Built a statistical simulation tool to estimate the probability for clones to fall within the top X% performers in CLD workflows.
• 📈 Used Monte Carlo simulation over synthetic/real assay data to model selection outcomes across multi-step processes.
• 🔁 Incorporated correlation-aware logic to simulate realistic assay dependencies and achieved ~10× speedup by using caching.
• 🖥️ Developed an interactive Streamlit dashboard for uploading Excel data, filtering, and visualizing assay distributions, clone probabilities, and histograms. Project Repository

• 🛰️ Built a hybrid Conv1D-RNN model along with YOLO for real-time object detection, tracking, and trajectory forecasting.
• ⚡ Achieved 5 ms/frame inference on NVIDIA Jetson Xavier NX for edge deployment.
• 📉 Conducted performance benchmarking against Kalman Filter, LSTM, and DeepSORT, achieving 20% RMSE reduction.
• 🏆 Focused on sequential modeling for flying objects (airplanes, UAVs) in real-time environments.

Open-Source NYC Computer Vision Hackathon with Moondream (Won 3rd Place)

• 🧠 Built a real-time AI system to detect, track, and identify unattended items in public spaces using the Moondream's VLM.
• 🌀 Designed a custom IoU-based object tracker with no external tracking libraries, enabling lightweight ID assignment.
• 🗣️ Leveraged Moondream's Vision language model for multimodal reasoning and image captioning to verify lost status and log item details in a searchable CSV.
• ⚡ Optimized for edge deployment with frame skipping, API caching, and fast video I/O for seamless real-time performance.
• 🔗 Project Repository

Reboot the Earth 2025 Hackathon - New York City

• 🌾 Developed an AI agent that helps farmers adapt to climate risks by predicting pest outbreaks, answering context-aware queries.
• 📡 Integrated Open-Meteo API for real-time forecasts; triggered alerts for rainfall, high winds, frost, and extreme heat.
• 🧠 Used RAG (Retrieval-Augmented Generation) not only for QA, but also for pest outbreak prediction by interpreting weather patterns against a knowledge base.
• 🗣️ Enabled dynamic, location-specific queries (e.g., “Will these conditions cause aphids?”) using curated NY-state pest articles.
  -🔗 AgroSynth-RAG Backend · Frontend Interface

• 🏞️ Developed an image captioning system using EfficientNet Encoder + T5-Small Decoder.
• 📈 Trained on Flickr8k / Flickr30k datasets, achieving strong BLEU-4 performance.
• 💾 Supports image summarization and question-answering over generated captions. Project Repository

• 🤖 Fine-tuned BERT-based genre classifier (F1-score: 0.92) on a dataset of 43k+ movies.
• 🧩 Combined semantic search with RAG (Retrieval-Augmented Generation) using LLaMA 3.2 for highly relevant movie suggestions.
• 💬 Built a Generative QA system to answer detailed queries about recommended movies (BLEU: 0.76).

• 📊 Architected a distributed ETL pipeline using Apache Kafka and Apache Spark for real-time cryptocurrency data ingestion and analysis.
• 🔮 Integrated LSTM-based forecasting with dynamic visualizations for five major cryptocurrencies (BTC, ETH, DOGE, XRP, SOL).
• 🧠 Incorporated sentiment analysis from live news feeds to enrich predictive modeling. Project Repository

• 💬 Engineered a scalable search engine on 150k+ tweets using PostgreSQL + MongoDB hybrid database.
• 🌍 Integrated language detection, machine translation, and sentiment analysis for content-aware search.
• ⚡ Reduced query latency by up to 4 seconds via LRU caching. Project Repository

• 📈 Engineered a predictive model to estimate news article popularity using AdaBoost, XGBoost, and Artificial Neural Networks (ANN).
• 📊 Analyzed a large dataset of 450K+ rows and 60 features, with the target variable (article shares) following a Poisson distribution.
• 🛠️ Applied advanced feature engineering, including log transformations and regularization techniques to handle skewed distributions and improve model robustness.
• 🧠 Optimized the neural network architecture to align with the Poisson-distributed target, achieving an MSE of 0.35, validated through bootstrap hypothesis testing.
• 🏆 Compared performance across models and validated findings using rigorous statistical testing and error analysis. Project Repository

Python | PyTorch | TensorFlow | Apache Kafka | Apache Spark | PostgreSQL | MongoDB | NVIDIA Jetson | Transformers | LLMs | Computer Vision | NLP | RAG | Streamlit | Docker | Git

• 📧 Email: [email protected]
• 🔗 LinkedIn
• 🌱 Open to collaborations, internships, and full-time opportunities.

"Building AI systems that don’t just work on paper but deliver results in the wild." 🧠🚀