Raj Tibarewala
AI / ML Engineer · Software Engineer — machine learning for autonomous systems
raj.tibarewala@gmail.com +91 91298 37866 github.com/RajTib linkedin.com/in/raj-tibarewala raj-tib.vercel.app Vellore, India
Education
Vellore Institute of Technology, Vellore
Jul 2024 – May 2028B.Tech Computer Science and Engineering (Cyber Security) · CGPA 8.65 / 10
Experience
Team Ardra — Software & R&D, Autonomous Drone Systems
Sep 2025 – Present- Build and validate embedded Linux software for onboard compute (NVIDIA Jetson Orin Nano, Raspberry Pi) on autonomous drone platforms.
- Prototyped an onboard vision-serving architecture (FastAPI, PyTorch, Hugging Face DETR) that decouples camera capture, threaded model inference and live video streaming, so slow inference does not stall the feed.
- Deploy edge ML inference within latency, power and compute budgets; debug integration issues across sensor, compute and control layers.
Pinnacle Labs — Cybersecurity Intern (remote)
Feb – Mar 2026- Built Python security tooling: a hybrid AES + RSA encryption tool and an input-capture research tool for studying interception techniques.
Projects
AegisFlight — ML-Based Drone Intrusion Detection
2026 · OngoingIIT Bombay TechFest 2026, PUSHPAK Grand Challenge (Stage 1 submitted) · Python, scikit-learn, NumPy, Pandas, pymavlink, FastAPI, React
- Explainable IDS over simulated MAVLink 2 telemetry, fusing protocol-rule, physics-consistency, ML anomaly (benign-only Isolation Forest + Mahalanobis over 11 online features) and SHA-256 firmware-integrity detectors.
- Reproducible evaluation: session-level splits, multi-seed attack-scenario benchmark (23k+ decisions), ML-off ablation, and replay of public PX4 / ALFA flight logs for false-alarm and sim-to-real (KS test) analysis.
- FastAPI / WebSocket backend, React dashboard, and a pytest suite spanning unit, integration, end-to-end and API tests.
GeoAI — Rooftop Mapping from Drone Orthophotos
2025 · FinalistIIT Bombay TechFest 2025 · Python, Ultralytics YOLOv8, SegFormer, OpenCV, rasterio, Shapely, Docker
- Turned village-scale GeoTIFF orthophotos and shapefiles into training data: CRS alignment, 640×640 tiling, empty-tile filtering, shapefile-to-label generation, 70/20/10 split, label and CRS validation, visual QA.
- YOLOv8 rooftop detection with four-class roof classification (RCC, Tin, Tiled, Others) and utility detection; SegFormer segmentation of water bodies and roads; Dockerized inference.
India Space Academy · Python, NumPy, Pandas, Matplotlib
- Estimated the Hubble constant and the age of the universe from Type Ia supernova data; compared the result against Planck 2018.
Python, OpenCV, MediaPipe
- Real-time computer-vision pipeline mapping webcam-detected hand gestures to OS-level file operations.
Research
“Nuclear Thermal Rocket: Design Considerations and Efficiency Optimization for Deep-Space Explorations”
Co-authored under faculty supervision — literature review, background research, technical structuring and manuscript preparation.
Skills
- Languages
- Python, C++, C, Java, JavaScript, SQL
- ML & data
- scikit-learn, NumPy, Pandas, SciPy, Matplotlib, Jupyter; anomaly detection, feature engineering, model evaluation, ablation, statistical analysis
- Deep learning & CV
- PyTorch, Ultralytics YOLO, Hugging Face Transformers, OpenCV; object detection, semantic segmentation, geospatial imagery (rasterio, Shapely)
- Systems
- FastAPI, Docker, Linux, Git / GitHub, pytest, NVIDIA Jetson Orin Nano, Raspberry Pi
- Frontend
- React, JavaScript
- Learning
- LLMs, RAG, LangChain, inference optimization