I'm an undergraduate Computer Engineering student at Kathmandu University, Nepal. I focus on software engineering, with a growing interest in applying machine learning and NLP to build things people can actually use.

pranaya:
  education: B.E. Computer Engineering, Kathmandu University
  interests: [Web Development, NLP, Deep Learning]
  currently_building: SignBridge - a real-time ASL translation platform
  open_to: [Software Engineering Roles, Internships, Open Source Collaboration]

_{placeholder name — swap in your actual repo name and link once it's public}

This platform bridges the communication gap between written/spoken English and American Sign Language. It combines a hybrid deep learning model for real-time webcam gesture translation, a rule-based English-to-ASL grammar glosser, and a Next.js 15 dashboard with gamified learning, custom quizzes, and analytics.

View Repository →

Key Features

1. Real-Time ASL Video Translation

• Landmark Tracking — tracks user movement using MediaPipe Holistic (11 pose points, 21 left-hand points, 21 right-hand points).
• Deep Learning Engine — translates 30-frame sequence streams into dynamic word classifications using a hybrid Conv1D + BiLSTM + Multi-Head Self-Attention model.
• Performance Optimization — applies feature scaling and normalization based on training-set IQR and medians for resilient, low-latency predictions.

2. English-to-ASL Text Glossing

• ASL Grammar Engine — restructures written English into standard ASL syntax order (TIME – TOPIC – COMMENT – WH).
• NLP Pipeline — uses NLTK tokenizers and lemmatizers to resolve contractions, handle spatial index mapping (this, that), replace pronouns, and isolate instrumental phrases.
• Video Playback — maps parsed tokens onto the WLASL dataset's video clips, rendering translations as a continuous sequence.

3. Gamified Study Sandbox

• Dynamic Flashcards — dual-sided cards (word on front, ASL demo video on back) with bookmarking, learned flags, and feedback.
• Adaptive Quizzes — a 10-level progressive quiz system with randomly generated video-based distractors.
• Daily Practice & Scheduler — generates 5 adaptive daily cards based on current level, tracks consistency, and supports custom study reminders.

4. Analytics & Progress Visualization

• Interactive Skill Trees — ReactFlow graphs mapping levels and concept tags.
• Skill Radar — radar chart across Knowledge, Practice, Accuracy, Satisfaction, Quizzes, Completion, and Engagement.
• Consistency Heatmap — visualizes daily activity over a 30-day period.

Technical Architecture

  [ USER TEXT INPUT ]                    [ WEBCAM INPUT FRAME ]
          │                                        │
          ▼                                        ▼
   ( NLTK POS Tagger )                    ( MediaPipe Holistic )
          │                                        │
   ( Lemmatizer (WordNet) )                        ▼
          │                            [ Extract Core Keypoints ]
   ( ASL Grammar Resolver )               • 11 Pose Coordinates
   - Time-Topic-Comment-Wh                • 21 Left Hand Coordinates
          │                               • 21 Right Hand Coordinates
          ▼                                        │
   [ Mapping Dictionary ]                          ▼
   - WLASL Video Database                 [ Robust Scaling (IQR) ]
          │                                        │
          ▼                                        ▼
 ┌──────────────────┐                     [ Hybrid Deep Network ]
 │ Next.js Sequence │                     • Conv1D Pathway
 │   Video Player   │                     • Bi-LSTM Sequential Pathway
 └──────────────────┘                     • Multi-Head Self-Attention
                                                   │
                                                   ▼
                                          [ Softmax Classification ]

_{self-rated — adjust the percentages to match your own level}

Feel free to reach out — always happy to talk about software, NLP, or potential opportunities.