I build end-to-end machine learning systems for space tech and healthcare — the kind that run on real scientific data and get judged on blind hold-outs, not notebook accuracy.

Most of my work lives at the messy end: raw instrument data, brutal class imbalance, and metrics that have to survive an audit.

_{B.TECH AI/ML  ·  RAIT, DY PATIL UNIVERSITY  ·  CLASS OF '27}

_{upper management}

EOS 6  

_{TASK-ROUTED MIXTURE OF EXPERTS}

Hand it raw solar X-ray telemetry and it does the rest — builds its own features, routes them through every expert, and streams one combined result: what is flaring right now, and the probability of a major flare in the next 24 hours.

LUMEN 6*  

_{P(MAJOR FLARE, NEXT 24H)}

Gradient-boosted ensemble over magnetogram-derived active-region features, with calibrated probabilities.

SPECTRA 6*  

_{LIVE FLARE STATE + CLASS}

Twin temporal convolutional networks reading per-minute X-ray telemetry, one per channel.

ROENTGEN

_{solar flare nowcast + 24h forecast}

Mixture-of-experts over dual-channel solar X-ray telemetry, soft and hard: two TCN nowcasters, a gradient-boosted forecaster, and an independently built flare catalogue. Above the published attention-based SOTA on SWAN-SF.

MAGNUSON

_{geomagnetic storm onset forecasting}

1M+ 10-second magnetometer observations from L1 orbit driving a real-time onset pipeline, with seven sequential architectures benchmarked head to head.

CME DETECTION

_{autonomous coronal mass ejection detection}

Two-model TCN + TCAN ensemble over in-situ solar wind plasma at L1 — bulk velocity, proton density, thermal temperature, helium-to-proton ratio and their gradients. Causal dilated convolutions give a 10.6-hour receptive field, so the model only ever looks backwards and stays valid for real-time use.

Beat XGBoost (0.210) and BiLSTM (0.195) on identical data. F1 sits near the physical ceiling for single-point plasma sensing — 30 to 50% of CMEs are stealth events that leave no plasma precursor at all. On the unseen May 2026 eruption it peaked at P = 0.8707, corroborated by a helium-to-proton ratio of 0.2965, roughly 7x the quiet-wind baseline.

SPACE WEATHER CLI

_{live interplanetary telemetry}

Async CLI aggregating NOAA and NASA DONKI feeds, built for headless systems with zero graphical dependencies.

NEUROPHARMA

_{drug-drug interaction risk engine}

Cross-references patient prescription profiles against pharmacological datasets and generates automated hazard alerts for clinical review. Next.js on Vercel, FastAPI and Docker on Hugging Face.