class PushpakSivaSai:
role = "AI/ML Engineer & Robotics Student"
education = "B.Tech CSE (AI) @ Amrita Vishwa Vidyapeetham"
cgpa = "8.25 / 10.0"
focus_areas = [
"Computer Vision & Markerless Motion Capture",
"Dual-Camera 3D Skeletal Triangulation & Kinematics",
"Edge-AI (Raspberry Pi 5 / Edge TPU) & ICS Security",
"Acoustic Anomaly Detection & Bio-ML / Quantum ML"
]
def how_i_engineer(self):
return (
"Understand the underlying mathematics, "
"build resource-efficient ML pipelines, "
"and deploy real-world perception systems."
) |
|
Apr 2026 – Jul 2026 | Guided by Dr. Akhil V. M. | Project Code
- Developed MoCap Studio, an end-to-end, local-first markerless motion capture system for video-based human movement analysis using MediaPipe Pose, Python, computer vision, and 3D reconstruction.
- Engineered an offline video-processing pipeline with FastAPI, enabling frame-wise human pose tracking and calibration-aware 3D skeletal reconstruction via dual-camera triangulation.
- Extracted biomechanical and kinematic metrics from human motion, exporting structured data formats (CSV, JSON, annotated video, 3D motion files).
2026 | sri-vartali-commerce
- Developed a complete full-stack e-commerce web platform for Sree Vartali Sarees with a responsive customer storefront, REST APIs, database integration, product management, and customer authentication.
- Maintained reliable data flow and API integration across frontend and backend services following an iterative production SDLC.
- Shipped 15 merged production pull requests (Phases 11–21) covering admin dashboards & a Super Admin panel, schema-driven faceted search, WhatsApp-first checkout, branding & splash screens, and production acceptance audits.
📦 All 15 merged pull requests — Phases 11–21 (click to expand)
| Pull Request | Deliverable |
|---|---|
| #21 · Phase 21 | Official company logo from the owner-provided artwork |
| #20 · Phase 20 | Company logo everywhere, splash screen, account registration |
| #19 · Phase 19 | Elegant bag cart icon, sarees-focused navbar, password-first sign-in |
| #18 · Phase 18 | Center navigation section |
| Spec v3.1 | Build spec v3.1 — recorded Phases 12–17 and the owner-directed deviations |
| #17 · Phase 17 | Owner UX pass — active nav, Buy-via-WhatsApp on product pages, 1440px layout, logo |
| Hotfix | Tolerated browser-extension attribute injection on html/body (hydration warning) |
| #16 · Phase 16 | WhatsApp-first commerce — cart Buy-via-WhatsApp + persistent supplier contact dock |
| #15 · Phase 15 | Production smoke runbook |
| #14 · Phase 14 | §58/§59 acceptance audit — order-ID search + touch-friendly image reorder |
| #13 · Phase 13 | Schema-driven facets, collection filter, collection-name search |
| Docs | Pointed the Phase 12 runbook at the owner's env location |
| #12 · Phase 12 | Super Admin panel (§9) + dashboard realtime (§39) |
| Phase 11b | Phase 11 completion — missing §47/§48 mobile + image-performance items |
| Dashboards | Client dashboard — categories, collections, inventory, customers & shipping management pages |
🛡️ Secure Smart Grid Industrial Control Systems: Cryptographic Encryption, Anomaly Classification, and Zero-Day Attack Detection
Jul 2026 | Iranian Journal of Science and Technology, Transactions of Electrical Engineering | Repository
- Engineered a real-time Edge-AI cybersecurity platform for Industrial Control Systems (ICS) using Google Coral Edge TPU.
- Combined machine learning-based intrusion detection with hybrid cryptographic protection (ECC + HKDF) for smart grid payloads and zero-day attack mitigation.
🪵 Hybrid Latent-Space and Reconstruction Scoring for Edge Acoustic Anomaly Detection in Industrial Wood Planers
2026 | Accepted for Publication
- Proposed a hybrid acoustic anomaly detection framework combining latent-space One-Class SVM scoring and Autoencoder reconstruction error for industrial machine condition monitoring.
- Built an Edge-AI inference pipeline deployed on Raspberry Pi 5 using PCA, ONNX Runtime, and residual autoencoders, achieving 90.48% anomaly recall.
I actively contribute upstream to leading open-source machine learning, security, and developer ecosystems — 8 merged pull requests (and counting):
| Ecosystem / Lab | Contribution Summary | Pull Request | State |
|---|---|---|---|
| Meta ThreatExchange | Tolerated empty list serialization for caseNumbers in STOPNCII threat data |
#2015 | 🟣 Merged |
| Meta Astryx | Fixed DropdownMenu submenu flyout auto-closing behavior on light dismiss |
#6912 | 🟣 Merged |
| Microsoft PyRIT | Preserved characters that the ASCII art font cannot render in console output | #2962 | 🟣 Merged |
| Repowise | Joined the stderr drain before building the doctor crash detail | #3165 | 🟣 Merged |
| Repowise | Deduplicated Rust mod items per #[path] attribute in code ingestion pipeline |
#2943 | 🟣 Merged |
| Repowise | Suppressed defer_in_loop health checks when subsequent statements break out of loop |
#2942 | 🟣 Merged |
| ONEARMY Community Platform | Migrated ConfirmModal to the new UI component system |
#4942 | 🟣 Merged |
| Kana Dojo | Added a new Japanese proverb to the learning content library | #31350 | 🟣 Merged |
| Microsoft Semantic Kernel | Excluded ClassVar type annotations from generated Python JSON schemas |
#14528 | 🟢 Open |
| Keras | Untracked sub-layers correctly in Layer.__delattr__ to prevent memory retention |
#23781 | 🟢 Open |
| Project MONAI | Informative logging for missing image-writer packages when ImageWriter fallback occurs |
#9142 | 🟢 Open |
| Project & Repository | Domain / Focus | Key Technical Approach & Achievements | Stack |
|---|---|---|---|
| 📷 MoCap Studio | Computer Vision & Robotics | Local-first multi-camera markerless 3D motion capture pipeline. Features frame-wise pose tracking with MediaPipe, calibration-aware dual-camera DLT triangulation, 3D skeletal reconstruction, kinematic metrics export, and 3D avatar visualization. | Python MediaPipe FastAPI OpenCV Panda3D Three.js |
| 🛡️ Smart-Grid ICS Security | Edge-AI & Cybersecurity | Journal publication implementation. Real-time Edge-AI cybersecurity platform integrating hybrid ECC + HKDF payload encryption with ML anomaly detection and zero-day attack classification on DNP3 Industrial Control System traffic. | Python TensorFlow Lite Cryptography Edge TPU scikit-learn |
| 📱 projX – OMR Evaluation App | Mobile Vision & Document AI | Offline Android application for automated OMR sheet detection and grading using smartphone cameras. Built fully on-device bubble detection with OpenCV, versioned answer key evaluation, and PDF report compilation. | Flutter OpenCV SQLite Dart |
| 🧬 MicroKPNN-MT | Bio-ML & Quantum Computing | Hybrid Quantum Knowledge-Primed Multi-Task Neural Network for microbiome-based disease prediction using biologically informed taxonomy-masked layers, multi-task auxiliary heads, and quantum residual fusion evaluated against SVM/RF/XGBoost. | PyTorch Quantum ML Bioinformatics XGBoost Captum |
| 👗 StyleSense FashionAI | Multimodal CV & LLMs | AI-powered personalized fashion recommendation platform. Extracts facial/body attributes (skin tone, eye color, hair color, body shape) via MediaPipe, drives XGBoost recommendation models, and features an interactive Groq LLM stylist chatbot. | Python MediaPipe XGBoost Groq LLM Streamlit |
| 🧠 EEG Signal Reconstruction | Biomedical Signal Processing | EEG signal recovery framework applying the Alternating Direction Method of Multipliers (ADMM) and sparse optimization techniques to reconstruct high-quality neural signals from noisy or incomplete channel measurements. | MATLAB ADMM SVD Signal Processing |
| 🧪 Cancer Driver Gene Classifier | Bioinformatics ML | Machine learning classification pipeline for cancer driver gene identification, incorporating genomic feature engineering, model benchmarking, and interpretable classification workflows. | Python Machine Learning Genomics scikit-learn |
| 🤝 HireHub Platform | Full-Stack Web App | Web platform connecting recruiters directly with student candidates for internship discovery, profile matching, and end-to-end application management. | React Node.js Express MongoDB |
| 🌾 KISSAN-BAZAAR | Full-Stack E-Commerce | Agriculture marketplace web application applying object-oriented design and core data structures. Features product catalog, shopping cart, authentication, and role-based management dashboards. | React TypeScript Supabase TailwindCSS Vite |
| 🎮 Double DQN Agent | Reinforcement Learning | Deep Reinforcement Learning implementation of a Double Deep Q-Network (Double DQN) agent trained on Tic-Tac-Toe using experience replay, target networks, and validation against a minimax baseline. | Python PyTorch RL NumPy |
| Category | Technologies & Tools |
|---|---|
| Programming Languages | Python · C/C++ · Java · JavaScript · TypeScript · SQL · R · MATLAB · Dart |
| AI / Machine Learning | PyTorch · TensorFlow · TensorFlow Lite · scikit-learn · Hugging Face · XGBoost · Captum · Quantum ML |
| Computer Vision & Perception | MediaPipe · OpenCV · Panda3D · Three.js · Human Pose Estimation · 3D Triangulation · Kinematics |
| Web & Mobile Development | FastAPI · Flask · Flutter · Next.js · React.js · Express.js · Node.js · REST APIs |
| Databases & Analytics | MySQL · MongoDB · Supabase · SQLite · Pandas · NumPy · SciPy · Power BI · Matplotlib |
| Hardware & Systems | Raspberry Pi 5 · Google Coral Edge TPU · ESP32 · ZeroMQ · POSIX Syscalls · CUDA · ONNX Runtime |
| Developer Tools | Git · GitHub · Linux / Bash · VS Code · Conda |
|
|
I am actively seeking AI/ML, Computer Vision, & Robotics Internship opportunities, research collaborations, and open-source challenges.
