DAchievements
Competing in public.
Two competitions, two very different systems: an autonomous underwater vehicle and a real-time transit assistant.
Singapore AUV Challenge
6th place
6th place globally · 320+ teams
- teams entered
- 320+
- teams qualified
- 130+
- teams shortlisted
- 20
- final placing
- 6th
Each dot is one team in a field of 320+: tinted dots mark the 130+ that qualified, bright dots the 20 shortlisted, and the large dot 6th place. Funnel bars use a square-root scale. Illustrative.
Autonomous underwater vehicle
An international student competition in which autonomous underwater vehicles complete navigation and manipulation tasks in a pool without human control.
Photos · tap a photo to view it full size
- From a field of 320+ teams, the team was among 130+ that qualified and 20 that were shortlisted, finishing 6th overall.
- Built autonomous navigation and object detection for the vehicle while at the Sri Sairam Techno Incubator Foundation.
- Trained YOLOv5 detectors with OpenCV for real-time underwater perception.
- Developed reinforcement-learning control on ROS, validated in Gazebo and in physical pool tests with a 92% success rate on obstacle avoidance, gate traversal and object recovery.
- YOLOv5
- OpenCV
- Reinforcement learning
- ROS
- Gazebo
Evidence
- LinkedIn post ↗Sree's post about the result
- Official competition site ↗sauvc.org
- Resume ↗Listed under Machine Learning Engineer, 2022
Podium graphic, illustrative.
Charlie: MBTA Real-Time Chatbot
Charlie answers live Boston transit questions by combining real-time MBTA data pipelines, delay-prediction features and a retrieval-augmented assistant.
Demonstrating Charlie, the MBTA real-time chatbot · tap a photo to view it full size
- Real-time ETL with Python, Apache Airflow, PySpark and Azure turns live MBTA data into features for transit delay prediction.
- Dockerized Azure ML services with autoscaling and automated retraining; Prometheus and Slack alerts surface service failures.
- A GPT-powered RAG assistant answers live transit queries; retrieval evaluation reached 92% precision on route and schedule questions.
- Python
- Apache Airflow
- PySpark
- Azure ML
- Docker
- Prometheus
- RAG
Evidence
- LinkedIn post ↗Sree's post about the 3rd-place finish
- Resume ↗Listed under Projects
- Repository ↗Placeholder repository; source code is not yet published
