Sandeep Das

Artificial Intelligence & Robotics MSc student · Technische Universität Nürnberg (UTN)
Building and evaluating machine-learning systems in computer vision, robotics, geospatial analytics and applied AI.

Portfolio · LinkedIn · Repositories

I work across the full experimental lifecycle: data pipelines, baselines, model training, stress testing, reproducibility and usable interfaces. I prefer documented trade-offs and measured results to unsupported performance claims.

Selected work

Project	Engineering or research focus
SignBridge-ASL	American Sign Language recognition research across static images, isolated video, reliability and evaluation; no verified public model release yet
Deepfake robustness benchmark	Deepfake detection under compression, resizing and blur; video-grouped evaluation, calibration and uncertainty
Sim2Real-Bench	Panda manipulation in MuJoCo/robosuite, policy robustness, failure prediction and reproducible evaluation
Nuremberg land-cover intelligence	Sentinel-2, 100 m land-cover estimates, spatial cross-validation and an interactive analysis tool
European image geolocation	Compact hierarchical visual geolocation and spherical decoding
Cabin Atlas	Aircraft-aware passenger/crew service application, transactional inventory and explainable recommendations
Voice Comparison Studio	Reference-based comparison of multiple speech-synthesis engines with a separate FastAPI backend
Redshift streaming analytics	Streaming data ingestion and analytics workflows

Browse all repositories →

Research and measured outcomes

• Land-cover composition: built-up fraction, five-fold spatial CV: MAE 0.0489 ± 0.0029, RMSE 0.0979 ± 0.0050, R² 0.9146 ± 0.0091. Details in the land-cover project.
• Image geolocation: reported 150.67 km median validation error for a 4.63M-parameter model; see experiment details.
• ASL research: co-authored Deep Gesture Interpretation for American Sign Language (2024). The newer SignBridge repository is an evolving research implementation; its current code is not an independently validated deployment.

Project-level READMEs distinguish reported experiments, tests, prototypes and work still requiring external validation.

Toolkit

Languages: Python, C/C++, TypeScript, JavaScript, SQL
ML and vision: PyTorch, TensorFlow/Keras, scikit-learn, LightGBM, OpenCV, EfficientNet
Robotics: ROS 2, MuJoCo, NVIDIA Isaac Sim/Lab, Webots
AI applications: Hugging Face, retrieval-augmented generation, LangChain, FAISS, MCP
Data and delivery: PostgreSQL, DuckDB, Kafka, Docker, FastAPI, React, GitHub Actions

Background

I am pursuing an MSc in Artificial Intelligence and Robotics at UTN in Nuremberg, Germany. Previously, as an AI & Automation R&D intern at ICAR–NBSS&LUP, I built Python APIs and automated data-access workflows.

I am interested in applied AI, robotics, computer vision and research engineering opportunities. The best entry points are the project READMEs, which contain the actual scope, setup instructions, evaluation methods and limitations.

Contact: LinkedIn · Portfolio