Graduate student and software engineer working on Artificial Intelligence and data-driven systems — with a particular interest in whether a model's explanation actually reflects what the model did. I like projects that end in a number I can defend.
📍 Houston, TX · 🌐 Portfolio · 📫 ashrith.vadde@gmail.com
- A Real-Time Human Computer Interaction Using Hand Gestures in OpenCV — ICTIS 2023, Springer, Lecture Notes in Networks and Systems vol. 720, pp. 271–282. DOI: 10.1007/978-981-99-3761-5_26
- Playing Maze using Voice Commands — IEEE ICCCNT 2023, Delhi, India, pp. 1–7. DOI: 10.1109/ICCCNT56998.2023.10307127
Explainability & NLP
| Project | What it does |
|---|---|
| CFPB Complaint XAI Audit | Fine-tuned DistilBERT for consumer-complaint routing (macro-F1 0.85), plus a quantitative faithfulness audit: Integrated Gradients, SHAP and attention rollout agree on only 13–29% of their top-attributed tokens |
| Sentence Embeddings for Summarization | Five encoders benchmarked under LexRank on 2,225 BBC articles; LaBSE leads on ROUGE-1/2/L |
| ApplyTeX | LaTeX-native, evidence-grounded resume tailoring for ATS-focused applications |
Reinforcement Learning
| Project | What it does |
|---|---|
| 2048 with Deep RL | DQN vs Dueling DQN in PyTorch, tuned with Optuna; reaches the 1024 tile and beats the published baseline |
| RL Navigation | Q-Learning, Dueling Double DQN with PER, and PPO on a grid world with dynamic obstacles, benchmarked against Dijkstra |
| Atari Breakout with DQN | Deep Q-Learning agent trained from raw gameplay frames |
Computer Vision & Robotics
| Project | What it does |
|---|---|
| Hand-Gesture HCI | 📄 Virtual mouse and keyboard from webcam hand gestures — OpenCV + MediaPipe |
| Voice-Controlled Maze | 📄 CNN speech-command recogniser at 94.57% accuracy driving a MATLAB maze game |
| Lawn-Mowing Robot | Autonomous coverage planning and LiDAR obstacle avoidance in ROS + Gazebo |
| Face Filters | Snapchat-style filters with Haar cascades and a PyQt5 GUI |
📄 = published work
Languages · Python · MATLAB · Java · SQL
ML / DL · PyTorch · TensorFlow · Keras · scikit-learn · Hugging Face Transformers · PySpark MLlib
Focus areas · Explainable AI · NLP · Reinforcement Learning · Computer Vision · Robotics (ROS)
Tools · Git · Docker · Jupyter · Gazebo · OpenCV · MediaPipe
Working on explainability and faithfulness in NLP models, and on generative-AI applications. Open to collaborating on open-source ML tooling.