Skip to content
View akashnathai's full-sized avatar
🏠
Working from home
🏠
Working from home

Block or report akashnathai

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
akashnathai/README.md

Typing SVG



👋 About Me

I'm an AI engineer and researcher from Silchar, Assam, building AI systems that hold up under real-world constraints: noisy clinical data, low-resource languages, and hardware that fits in your palm. I care about models that are not just accurate, but explainable and honest about their uncertainty.

  • 🔭 Now: AI Researcher on the Indian Knowledge Systems (IKS) project at NIT Silchar, working on AI that classifies and generates Indian music
  • 🏆 Best Paper Award, AICTA 2026 (NIT Silchar, Springer LNNS) for BioSignal-LLM
  • 🎤 First-author oral + e-poster at WSDM 2025, Dehradun
  • 🎓 B.Tech in Computer Science & Engineering, Assam University (2021–2025)
  • ⭐ CodeChef 4★ competitive programmer
  • 📫 Open to research collaborations, reach me at akashnath.aus@gmail.com

🚀 Featured Projects

Project What it does Key result Status
🩺 MEDIAX Brain-tumor MRI classifier with uncertainty-aware triage (EfficientNet-B3, multi-scale fusion, MC Dropout) 93.75% acc · 0.983 AUC · glioma miss-rate 22% → 2.8% Under review, IEEE CVMI 2026
🫀 BioSignal-LLM Fuses 12-lead ECG with a LoRA-adapted BioGPT to classify cardiac conditions and explain its reasoning 0.874 macro AUC on PTB-XL 🏆 Best Paper, AICTA 2026
📜 Sanskrit GPT-2 97.7M-parameter Sanskrit language model with a custom 16K Devanagari BPE tokeniser, trained on AI4Bharat Sangraha — In progress
🖼️ SANSCAP Sanskrit image captioning with CNN-LSTM vision-language models — Research code
🔥 FireSense Real-time fire-severity classifier running on an ESP-32 CAM 94.39% acc on MIVIA · < 500 ms on-device Presented, WSDM 2025
🏔️ Himalaya EWS Explainable landslide & flash-flood early-warning map for the Eastern Himalaya (Sentinel-2, SRTM, IMD rainfall; SegFormer, U-Net++) 0.874 CV AUC on 420 landslides Presented, National Seminar 2026
🐢 TurtleNet Explainable turtle species classification with attention-enhanced EfficientNetB4 + ensembling 97.25% accuracy Published, IJEEI (Scopus)
🌍 Hybrid AQI XAI CNN-BiLSTM air-quality forecaster for Delhi, Mumbai, Kolkata & Chennai with SHAP explanations · project site R² = 0.987 (Delhi) Submitted
✨ VINO AI Full-stack AI SaaS for image restoration, recolouring and semantic search (Next.js, TypeScript, MongoDB, Stripe) — Shipped

📄 Publications & Talks

  • Multimodal Clinical Decision Support via LLM–Biomedical Signal Fusion with Explainable AI AICTA 2026, NIT Silchar · Springer LNNS · 🏆 Best Paper Award
  • TurtleNet: Explainable Turtle Species Classification Using Attention-Enhanced EfficientNetB4 and Ensemble Learning IJEEI, Vol. 14(2), pp. 533–547, 2026 · Paper
  • AI-Driven Multi-Modal Early Warning Framework for Landslide and Flash-Flood Risk in the Eastern Himalaya National Seminar 2026, Assam University, Silchar
  • Real-time fire-severity AI on ESP-32 hardware WSDM 2025, Graphic Era University, Dehradun · First-author oral + e-poster

📚 Full list on Google Scholar · ORCID


🔬 Research Interests

Explainable AI Medical Imaging Multimodal LLMs Uncertainty Estimation Sanskrit & Low-Resource NLP Music AI Edge AI Remote Sensing Time-Series Forecasting


🛠️ Tech Stack

AI / ML

Also: Keras · PEFT/LoRA · SHAP · Grad-CAM · NumPy · Pandas · Weights & Biases

Languages & Web

Cloud, DevOps & Hardware


📊 GitHub Activity


Pinned Loading

  1. hybrid_aqi_xai hybrid_aqi_xai Public

    Explainable CNN-BiLSTM with SHAP-based seasonal & diurnal temporal attribution for AQI forecasting across Delhi, Mumbai, Kolkata, and Chennai. Trained on CPCB hourly data (2015–2020).

    Jupyter Notebook

  2. SANSCAP SANSCAP Public