Juhil Modi

Juhil Modi

Computer Science undergraduate · Scaler School of Technology and BITS Pilani · Bengaluru, India

I'm in the third year of a full-time undergraduate programme at Scaler School of Technology in Bengaluru, and I'm earning a B.Sc. (Hons.) in Computer Science from BITS Pilani in parallel (CGPA 8.93/10). I work on machine learning for scientific data, mostly radio signals, satellite imagery and, more recently, collider open data.

Request CV juhilmodiatwork@gmail.com LinkedIn

Research interests

  • Anomaly detection. Label-free and weakly supervised searches, and the ways they can produce false signals.
  • Data-quality monitoring. Finding inconsistent records, instrument artefacts and pipeline faults before they reach a result.
  • Signal search in radio data. Narrowband and periodicity search in filterbank and LOFAR data, with radio-frequency-interference (RFI) context.
  • Earth observation. SAR and multispectral pipelines, and geospatial foundation-model embeddings.

Projects

  • Anveshan Founder · private beta
    An Earth-observation platform that turns satellite data into reviewable, source-backed evidence.
  • ExoSignal Personal research · ongoing
    Narrowband and periodicity search in radio telescope data, with data-quality context on every candidate.
  • Collider open-data study 2026 · open source
    Dimuon resonances in CMS data and a label-free anomaly search on the LHC Olympics dataset. Summary · GitHub
  • DRISHTI Smart India Hackathon 2026 · team lead, team of 6
    Camera-only navigation for off-road vehicles without GPS. GitHub
  • Cropland mapping ITU GEOAI Challenge on Zindi · rank 5 · solo
    Crop classification from Sentinel-1 and Sentinel-2 time series.
  • Drone imagery and hydrology IIT Bombay Geo-AI Hackathon · 1st prize · team of 4
    Segmentation of drone orthophotos and drainage extraction from point clouds.
  • Sentinel / NETRA Team of 2
    Traffic-incident detection from CCTV and drone video, with human review.

Collider open-data study

A self-directed study on public particle-physics data. I rebuilt the J/ψ, Υ and Z peaks from CMS 2011 dimuon events, then ran a label-free dijet anomaly search on the LHC Olympics 2020 dataset. The plot compares four selection methods, and the crosses show the false excess each produces on background-only data. Code, results and a full report are on GitHub.

Significance in the signal region for no cut, autoencoder, CWoLa and a supervised upper bound, with background-only controls Significance in the signal region for no cut, autoencoder, CWoLa and a supervised upper bound, with background-only controls

Results and recognition

  • 1st Prize, National Geo-AI Hackathon, Techfest 2025–26, IIT BombayTheme 1 winner and Theme 2 finalist, in a team of four. Certificate
  • Rank 5, ITU GEOAI Challenge for Cropland Mapping in Dry Environments (Zindi, 2025)Final private leaderboard, 184 active participants, entered solo. Leaderboard
  • NASA ROSES-2025 C.12 (FAIMM) proposal, as sole Principal InvestigatorPeer reviewers rated its overall merit "Excellent". It was not selected for funding (August 2026).
  • Poster abstract accepted, 2026 Language AI in the Space Sciences Workshop, STScIRadio-spectrogram analysis with contextual retrieval from NASA ADS.

Contact

Email is the best way to reach me. The collider study and DRISHTI are open source; code for the other projects is available on request.