field logConference transmissions
👋 Hey, I'm Akash Nath.
I write research, ship products, and try to make deep learning work in places where people say it won't — on a $10 chip, in ancient scripts, inside an MRI scanner.
About
I grew up in Assam, Northeast India — a region where good internet arrives late and cutting-edge research arrives even later. That probably explains why I'm obsessed with building AI that works under constraints: on cheap hardware, in ancient languages, with limited data.
I started coding seriously at 17, I graduated from Assam University in year 2025, I have presented a research paper as first-author and deployed an AI model on an ESP-32 microcontroller costing less than a textbook. That project — an AI that classifies fire severity in real-time — was presented as an oral and e-poster at the World Summit on Disaster Management, 2025. It was my proof that serious research doesn't need a lab in a rich city.
Right now I'm building a brain-tumor MRI classifier that tells doctors when it's unsure. Most AI systems optimise for looking confident. I think a system that says "I don't know, please double-check this one" is far more valuable in medicine — so that's what I built. It's under review at IEEE CVMI 2026. Alongside it, I built a system that reads a 12-lead ECG and explains its diagnosis in plain clinical language — that one won the Best Paper Award at AICTA 2026. And because every monsoon the Eastern Himalaya loses hillsides with no warning, I built an explainable landslide early-warning map, which I presented at Assam University in July 2026.
I'm also training a GPT-2 for Sanskrit — because one of the world's oldest written languages deserves a language model, and nobody else in my corner of the world was doing it. When I'm not doing research, I'm shipping products, competing on CodeChef (4-star), and applying for M.Tech AI/ML programs.
Now
Most AI just gives you an answer. MEDIAX gives you an answer plus a confidence score — and when it's not sure about a glioma, it says so. That's the feature. Submitted to IEEE CVMI 2026.
Sanskrit is one of humanity's oldest knowledge systems and it has almost no AI tooling. I trained a 97.7M-parameter language model on the AI4Bharat Sangraha corpus with a custom Devanagari BPE tokeniser. Evaluation ongoing.
BioSignal-LLM fuses raw 12-lead ECG with a language model: a 4.2M-parameter 1D ResNet encoder projected into a LoRA-adapted BioGPT. It classifies five cardiac superclasses and writes the clinical explanation. 72.6% accuracy, 0.874 macro AUC on 18,750 PTB-XL recordings. Won the Best Paper Award at AICTA 2026, NIT Silchar (Springer LNNS).
Hybrid AQI XAI: a CNN-BiLSTM trained on six years of hourly CPCB data for Delhi, Mumbai, Kolkata and Chennai, with SHAP attribution. R² 0.987 on Delhi, and it points at 08:00 IST as the hour that drives PM pollution — something a policymaker can act on.
Published work
EfficientNet-B3 with multi-scale feature fusion and Monte Carlo Dropout. Hits 93.75% accuracy and AUC 0.983. The real contribution: a clinical triage layer that drops glioma miss-rate from 22% to 2.8% by trading some recall for safety. Built for clinicians, not benchmarks.
A frozen 1D ResNet ECG encoder projected into a LoRA-adapted BioGPT, with a joint classification head and GradCAM++ saliency over the signal. Classifies five cardiac superclasses on PTB-XL (72.6% accuracy, 0.874 macro AUC, +5.6 pts over ECG-only) and generates the clinical explanation alongside.
All-India rainfall thresholds break down in a belt this wet, so the terrain that needs warnings most often has none. This fuses Sentinel-2 imagery, SRTM terrain and 25 years of IMD rainfall into one explainable risk map, validated on 420 documented landslides: 0.874 cross-validated AUC, with 79.5% of known landslides in the high / very-high zones. Every alert carries SHAP + Grad-CAM reasons a district officer can read. Presented at the National Seminar on “Role of the Himalaya in Achieving the Vision of Viksit Bharat 2047”, theme Disaster Risk Reduction through Environmental Governance.
An attention-enhanced EfficientNetB4 feature extractor feeding an ensemble of classifiers, with explainability on top — built for conservation work, where a wrong species call has consequences. 97.25% accuracy.
A hybrid EfficientNetB3 + BiLSTM that classifies six fire-severity levels from a live camera stream in under 500ms — deployed on ESP-32 hardware. 94.39% accuracy on the MIVIA dataset. Presented as first-author oral and e-poster at the World Summit on Disaster Management, Graphic Era University, 2025.
97.7M parameters, trained on the AI4Bharat Sangraha corpus. Custom 16K Devanagari BPE tokeniser. Evaluation across perplexity, type-token ratio, and script purity. Composite eval score 7.55/10 at checkpoint 83,000. Draft paper targeting ACL/EMNLP/LREC-COLING.
Selected builds
EfficientNetB3 + BiLSTM hybrid that classifies six fire severity levels from an ESP-32 CAM stream in under 500ms. Uses a Gemini-assisted + human-expert labelling pipeline on the MIVIA dataset. First-author at WSDM 2025 in Dehradun. This is the one that got me into research.
97.7M-parameter transformer trained on the AI4Bharat Sangraha corpus with a custom 16,000-token Devanagari BPE tokeniser. Trained iteratively across RTX 2060, T4, and A100 environments with a tuned bf16 + AdamW + NVMe-cached pipeline. One of the few Sanskrit LLMs built from first principles.
Built from scratch in PyTorch — multi-scale EfficientNet-B3 fusion, MC-Dropout, temperature scaling, two-phase training. The clinical triage layer is the real innovation: it cuts glioma miss-rate from 22% to 2.8% by refusing to make a confident call when the model is genuinely unsure. Submitted to IEEE CVMI 2026.
Image restoration, recolouring, and semantic search in one platform. Clerk for auth, Stripe for subscriptions, MongoDB for persistence, and a credit-based metering system that makes inference cost predictable. This is what "shipping" actually looks like — not just a Colab notebook.
Sentinel-2 imagery, SRTM terrain and 25 years of IMD rainfall fused into one explainable susceptibility map for the Eastern Himalaya. 0.874 AUC on 420 known landslides. A SegFormer segmenter lands within ~2 F1 points of U-Net++ with 7× fewer parameters, at 18 ms a patch on a single CPU thread.
Attention-enhanced EfficientNetB4 features into an ensemble of classifiers, with explainability so a conservationist can see what the model looked at. 97.25% accuracy. Published in IJEEI (Scopus), June 2026.
CNN-BiLSTM over six years of hourly CPCB readings for six pollutants across Delhi, Mumbai, Kolkata and Chennai. SHAP GradientExplainer shows which hours and pollutants drive each forecast. Delhi: R² 0.987, MAPE 5.99%.
Where I've been
Tools
Highlights
FAQs
Get in touch
Whether it's research, an M.Tech opening, a consulting chat, or just wanting to nerd out about AI — my inbox is open.
I'm based in Assam, India. I reply to emails. I'm also active on LinkedIn and GitHub — the links are all right here. If you're a researcher, recruiter, or founder building something in AI, I'd genuinely love to hear about it.
Email me →