CNN-LSTM hybrid classifying over-the-air signal modulations, with a Software-Defined Radio as transceiver.
PyTorch · CNN-LSTM · SDR (over-the-air capture)
Signals, machine learning & systems, from radio waveforms and speech to brains and neural nets.
Engineering undergrad (Year IV) at Thapathali Campus, IOE. Building reliable systems and applied ML for signals, speech, and the brain-adjacent.
Signals as a throughline across radios, speech, brains, and models.
Electronics, Communication & Information Engineering graduate from Thapathali Campus, IOE, with strengths in signal processing, networks, and AI / ML. Experienced backend developer and former ECAST president, focused on reliable systems and applied machine learning that survives contact with the real world.
The throughline is signals: radio waveforms classified over the air, speech isolated from a cocktail party, MRI volumes triaged, hand gestures decoded from magnetic fields, different sensors, same discipline.
Fifteen catalogued builds, grouped by series. Methods and results as noted.
CNN-LSTM hybrid classifying over-the-air signal modulations, with a Software-Defined Radio as transceiver.
PyTorch · CNN-LSTM · SDR (over-the-air capture)
Multimodal cocktail-party solver: spatial audio from a 4-microphone array fused with visual face cues to isolate a chosen speaker.
Deep multimodal net · 4-mic beamforming · audio-visual fusion
Four-class MRI triage (glioma, meningioma, pituitary, none) comparing a custom CNN, ResNet-18 from scratch, and DenseNet-121 transfer.
CNN · ResNet-18 · DenseNet-121 (ImageNet)
Defended image classifiers by deflecting pixels and denoising in the wavelet domain, preserving accuracy with zero retraining.
Wavelet-domain denoising · adversarial robustness
Self-supervised model that learns normal driving dynamics from dashcam video and flags risk via epistemic uncertainty, with no crash data needed.
Self-supervised video ML · epistemic uncertainty
Fine-tuned CodeLlama to translate natural-language prompts into ready-to-render Manim animation scenes.
CodeLlama fine-tune · Manim
★ Awarded one of the Best Departmental Projects · 2026
NSGA-II multi-objective scheduler with DQN/PPO agents as a hyper-heuristic selecting repair strategies each generation, timetabling 444 courses, 181 instructors and 67 rooms at Thapathali Campus. Greedy repair pipeline cuts hard violations 92 to 94% in one generation.
NSGA-II · DQN & PPO hyper-heuristic · vectorized eval (5 to 6.5×)
Backend for a campus-wide timetable platform: multi-role access, real-time sync, conflict detection, workload tracking, analytics.
Django REST Framework · PostgreSQL
Backend API for the official campus website, built and maintained with a team of developers and designers.
Django · DRF
Sign-language-to-voice glove sensing hand gestures through Hall-effect magnetic fields.
Hall-effect sensors · embedded C · speech synthesis
Wearable for the visually impaired: ultrasonic obstacle ranging plus camera-based object detection.
Ultrasonic sensors · camera CV · embedded
Library management system emulating blockchain from scratch in C++ for tamper-evident, transparent records.
C++ · custom blockchain
Backend for a mental-health support platform with healthcare-specific APIs.
Django · DRF
Social platform for developers to share projects and ideas.
Django · full-stack
One command, any language: a CLI that auto-detects the project's test framework across 11 languages and renders beautiful, unified output.
Rust · framework auto-detection · single static binary
Roles and society work, 2021 to present, as a raster of activity.
May 2025 to present
Self-directed
May 2024 to Apr 2026
Thapathali Campus, Kathmandu
Apr 2023 to Apr 2026
Thapathali Campus, Kathmandu
2021 to 2022
Fiverr
May 2022 to May 2025
ECAST, Thapathali Campus
Electronics & Computer Students Amidst Students of Thapathali
Degrees, fellowships, and the academic record.
May 2022 to present · Year IV
Thapathali Campus, IOE · Kathmandu, Nepal
Mar 2024 to Jan 2025
Fusemachines AI Fellowship · Kathmandu, Nepal
May 2025 to present
NPLCoder · Kathmandu, Nepal
2019 to 2021
Orchid Science College · Nepal
Peer-reviewed and accepted publications, reverse chronological.
Padhya, D., Maharjan, S., Adhikari, B., & Pokharel, I. R. (2026). “IsoNet: Spatially-aware audio-visual target speech extraction in complex acoustic environments.” Journal of Innovations in Engineering Education (JIEE), accepted · arXiv May 2026.
Padhya, D., Pant, J., Acharya, K., Maharjan, S., & Thakur, S. K. (2026). “Design and Implementation of a Multi-Purpose Low-Cost Hall-Effect Sensor Glove for Sign Language Recognition.” KEC Journal of Science and Engineering (KJSE), 10(1), 96 to 102, May 2026.
Padhya, D. et al. (2025). “CNN-LSTM Hybrid Architecture for Over-the-Air Automatic Modulation Classification Using SDR.” Journal of Innovations in Engineering Education (JIEE), 8(1), November 2025.
Languages, frameworks and tools, with bar height denoting depth of practice.
spoken · Nepali (native) · English (fluent) · Hindi (fluent)
Competitive results and recognitions, reverse chronological.
Thapathali Campus, IOE
Hyper-heuristic university course timetabling using RL (DQN/PPO) and genetic algorithms
Kathmandu University
IsoNet: multimodal audio-visual target speech extraction
China
Represented Nepal; workshops on global technology trends and leadership
Pulchowk Campus
Universal Hand Gesture Decoder, project demonstration
Thapathali Campus
Hall-effect sign-language-to-voice glove
Asian College
Blind Guidance System: ultrasonic ranging + object detection
Open to research, collaborations, and good problems.