BTech student in AI and Data Science at Woxsen University. Seeking research internships in AI/ML.
Driven by the simple curiosity of how things work. Currently navigating the ever-morphing field of AI, one project at a time. Focused on creating technically sound solutions that bridge the gap between abstract intelligence and practical application.
Ongoing research in machine learning and applied AI. Published on Zenodo.
Status: Published on Zenodo
Exploring how multi-path Transformer architectures can induce complexity-adaptive, 'dual-system' reasoning without explicit supervision. THMI processes inputs through three parallel reasoning paths - shallow (System 1), medium (System 2a), and deep (System 2b) - with a confidence-weighted ensemble that learns to self-allocate reasoning depth. Achieves 98.58% accuracy vs. 91.06% baseline.
Read Deep DiveStatus: In Progress
A training-free safety framework that uses physics-informed latent-space dynamics to detect harmful prompts with negligible computational overhead.
Developed a first-principles AI pipeline that fuses Deep Learning (DenseNet121) with Physics-Informed Neural Networks (PINNs) to classify brain tumors with 96% accuracy and simulate their biological growth dynamics using reaction-diffusion PDEs.
Technologies: Deep Learning, PINNs, RAG, Mistral LLM, Medical AI
View on GitHubEngineered a dual-purpose astrophysical framework achieving 99.68% test accuracy in detecting planetary transits using LSTM networks and classifying habitability through a hybrid CNN-LSTM architecture trained on NASA Kepler/TESS data.
Technologies: LSTM, CNN, NASA Data, Space Science
View on GitHubBuilt a high-precision financial forecasting model using LSTM neural networks that achieved an R-squared value of 0.95, integrating technical indicators (RSI, MACD) and a MySQL backend.
Technologies: LSTM, Finance, SQL
View on GitHubArchitected a scalable NLP pipeline capable of processing 1 million social media posts per day, achieving 85% classification accuracy.
Technologies: NLP, PRAW, TextBlob
View on GitHubRAG, PINN, Neural networks, LLM Finetuning, ML, Computer Vision, Transformer Architecting, Neural Architecture Synthesis
Python, TensorFlow, PyTorch, scikit-learn, NumPy, Flask
MySQL, Hadoop, Power BI, Google Cloud, Docker, Git, Linux
Email: rohannambiar370@gmail.com
GitHub: github.com/Rohnnam
LinkedIn: linkedin.com/in/rohannam