Shubham Parida
Software Engineer & Builder
Building software that solves real-world problems.
I specialize in backend systems, real-time applications, and distributed architecture. From building embedded IoT networks to integrating machine learning into scalable software products, I focus on delivering robust and efficient solutions.
Experience
Intern, Software Development
RateGain Travel Technologies Limited
Noida, India · Jul 2026 – Aug 2026
- check_circle Developed and maintained Apache Airflow DAGs on Google Cloud Composer, orchestrating data pipelines integrated with BigQuery and Cloud Storage.
- check_circle Optimized the alerting system for failed Airflow DAGs, redesigning failure detection, processing, and notification logic to improve pipeline reliability and reduce manual triage.
- check_circle Integrated Jenkins into the DAG failure alerting workflow and debugged production-style pipeline failures across Airflow DAG logic and GCP service integrations, supporting containerized deployment with Docker.
Freelance Academic LaTeX Consultant
Self-employed · 5 college faculty clients · Apr 2025 – Present
- check_circle Automated LaTeX pipelines for IEEE, Springer, and Elsevier templates across 5 faculty clients, including Beamer decks; reduced typesetting turnaround by ~60% per manuscript vs. manual Word conversion.
- check_circle Structured complex manuscript elements — BibTeX/BibLaTeX bibliographies, algorithm environments, multi-column layouts, and figure/table formatting — for journal and conference submissions.
Projects
Click any project to expand details, tech stack, and links.
Research-driven & engineering-focused
Overview
A multiplayer location-based survival game where player interactions happen according to real GPS positions. Built a GPS proximity detection engine using the Haversine formula for accurate player collision and zone detection. Designed real-time multiplayer synchronization achieving sub-second latency for multiple concurrent players.
Key Features
- check_circle GPS proximity engine using Haversine formula for accurate geofencing
- check_circle Real-time multiplayer sync via self-hosted WebSocket backend
- check_circle Live zone entry/exit events & player elimination system
- check_circle Sub-second synchronization for multiple concurrent players
Tech Stack
Highlights
Overview
Developed a CNN + Sentence-BERT ensemble REST API classifying 13 emotion categories; deployed on Hugging Face Spaces with sub-250 ms inference latency and structured JSON response formatting. Evaluated the model with F1-score, precision/recall, and confusion matrix analysis; handled class-imbalanced dataset preparation using Pandas and NumPy with Matplotlib and Seaborn visualisations.
Key Features
- check_circle Developed a CNN + Sentence-BERT ensemble REST API classifying 13 emotion categories
- check_circle Deployed on Hugging Face Spaces with sub-250 ms inference latency & structured JSON response formatting
- check_circle Evaluated the model with F1-score, precision/recall, and confusion matrix analysis
- check_circle Handled class-imbalanced dataset preparation using Pandas and NumPy with Matplotlib and Seaborn visualisations
Tech Stack
Results
Overview
Engineered Selenium + BeautifulSoup scraping pipelines tracking trend signals across social and marketplace APIs; processed 10k+ daily data points into structured JSON/CSV with AJAX and infinite-scroll handling. Designed a sentiment modeling layer mapping scraped signals to directional prompts for Stable Diffusion img2img transformations; achieved real-time visual style evolution driven by live market trend data.
Key Features
- check_circle Engineered Selenium + BeautifulSoup scraping pipelines tracking trend signals across social & marketplace APIs
- check_circle Processed 10k+ daily data points into structured JSON/CSV with AJAX and infinite-scroll handling
- check_circle Designed sentiment modeling layer mapping scraped signals to directional prompts for Stable Diffusion img2img transformations
- check_circle Achieved real-time visual style evolution driven by live market trend data
Tech Stack
Scale
Overview
Implemented a CV-based adaptive traffic control system on Raspberry Pi + IoT hardware; achieved 25–35% reduction in congestion and 20–30% decrease in average junction wait times. Published in GIS Science Journal (Vol. 11, Issue 4, 2024); covered by The Times of India — research conceived and completed independently prior to undergraduate enrollment.
Key Features
- check_circle Implemented a CV-based adaptive traffic control system on Raspberry Pi + IoT hardware
- check_circle Achieved 25–35% reduction in congestion and 20–30% decrease in average junction wait times
- check_circle Published in GIS Science Journal (Vol. 11, Issue 4, 2024) · Media coverage by The Times of India
- check_circle Conceived and completed independently prior to undergraduate enrollment
Tech Stack
Impact
More Projects & Experiments
Open SourceExplore active repositories, full backend implementations, and upcoming releases.
Publications
Peer-ReviewedScholarly research contributions, peer-reviewed journal papers, and academic citations.
Temporal Convolutional Classification of RVC-Synthesized Speech Using Pre-Trained Audio Embeddings
Abstract & Method
This paper presents a real-time method for detecting fake speech generated using Retrieval-based Voice Conversion (RVC). The approach leverages Google HEAR audio embeddings to extract meaningful acoustic representations from short audio segments. A sliding window technique is applied along the embedding dimension of each speech representation to capture local structure across neighboring embedding coefficients. Three deep learning models—CNN-LSTM, Temporal Convolutional Network (TCN), and TCN-LSTM—are benchmarked on the DEEP dataset containing real and AI-generated speech with added background noise. Experimental results show windowed embedding representations significantly enhance detection, with the TCN model achieving 96.85% accuracy and low inference latency suitable for real-time audio forensic streams.
@article{hota2026temporal,
title={Temporal Convolutional Classification of RVC-Synthesized Speech Using Pre-Trained Audio Embeddings},
author={Hota, Dipu and Barik, Ram Chandra and Behera, Niranjan and Nayak, Sanjib Kumar and Parida, Shubham},
year={2026},
note={DOI: incoming}
}R.A.P.T.O.R — Raspberry Pi Automated Traffic Optimization and Response
Abstract & Summary
This paper presents an adaptive, smart traffic control system utilizing computer vision on embedded IoT edge hardware (Raspberry Pi). By leveraging OpenCV for continuous vehicle density estimation across intersection approaches, the system dynamically calculates and modulates signal cycle times. Experimental deployment demonstrated a measured 25–35% reduction in vehicular congestion and a 20–30% decrease in average junction wait times compared to traditional static signal timing. The research was conceived, engineered, and completed independently prior to undergraduate studies.
National news coverage spotlighting pre-college engineering research and intelligent traffic management.
@article{parida2024raptor,
title={R.A.P.T.O.R: Raspberry Pi Automated Traffic Optimization and Response},
author={Parida, Shubham},
journal={GIS Science Journal},
volume={11},
number={4},
year={2024},
issn={1869-9391}
}Technical Skills
AI / Machine Learning
Languages
APIs & Backend
Cloud, Data & DevOps
Scraping & Pipelines
Embedded & IoT
Document & Typesetting
Achievements
Competition wins, recognition & publications
HackFest-26
Outperformed 200+ teams at a national-level hackathon. Signal-86 recognised for real-world deployment viability, sub-second WebSocket performance under load, and novel GPS game mechanic design.
- · Outperformed 200+ teams
- · National-level hackathon
- · Novel GPS game mechanic design
Youth Ideathon 2023
Secured 2nd place nationwide at IIT Delhi (AIR 2, Gold Medal). Project ranked among Top 1,000 submissions on the Think Startup YouTube playlist.
- · Secured 2nd place nationwide at IIT Delhi
- · AIR 2 · Gold Medal
- · Top 1,000 submissions on Think Startup YouTube
R.A.P.T.O.R — Published
Published in GIS Science Journal (Vol. 11, Issue 4, 2024). Adaptive traffic control system. Media coverage: The Times of India.
- · GIS Science Journal, Vol. 11, Issue 4, 2024
- · Adaptive traffic control system
- · Media coverage: The Times of India
Machine Learning Specialization
Covers: Supervised ML · Advanced Learning Algorithms · Unsupervised Learning · Recommenders · Reinforcement Learning
Let's Connect
Open to research fellowships, AI/ML collaborations, and interesting engineering challenges.
Location
Bhubaneswar, Odisha, India
University
Siksha ‘O’ Anusandhan University (ITER) · B.Tech CSE
Sep 2024 – May 2028 · CGPA 8.33
Current Status
Open to research fellowships & AI/ML collaborations