AI Software Engineer
Splashgain Technologies
Pune, India
Conversational SystemsAngularJSSQLHighcharts
- Built and deployed a production conversational system for Eklavvya and e-Pravesh, automating support for 500+ global customers and 5,000+ concurrent users on enterprise platforms processing 30M+ assessments across 15+ countries, reducing repetitive support requests.
- Redesigned AngularJS components and data-loading workflows for Eklavvya, improving UI responsiveness by 35% under peak examination traffic through efficient data loading and responsive rendering, keeping high-concurrency assessment sessions stable for simultaneous users.
- Implemented SQL stored procedures for backend retrieval and reporting workflows, aligning retrieval logic with frequently accessed examination datasets to improve database access efficiency and response times for assessment reporting and administrative operations.
- Developed Highcharts monitoring dashboards for grades, comparative rankings, and student-performance analytics across 30M+ assessments, consolidating examination results into interactive real-time views that accelerated access to operational insights.
Software Intern
Splashgain Technologies
Pune, India
TensorFlowOpenCVEfficientDet-D0Computer VisionChatbots
- Engineered a real-time OpenCV computer-vision pipeline processing webcam streams at 20+ FPS to detect unauthorized phones and earphones during online examinations, automating proctoring without interrupting live sessions across varied webcam hardware.
- Fine-tuned an EfficientDet-D0 object-detection model with TensorFlow on custom annotated datasets, validating predictions across varied lighting conditions and webcam hardware to reliably identify unauthorized objects during real-time examination proctoring.
- Analyzed recurring support-query patterns and built an automated chatbot workflow for common user requests, translating repeated request patterns into automated flows that reduced manual responses and laid the foundation for the production conversational system.
Data Analytics Intern
Mahindra and Mahindra
Nashik, India
PythonPandasNumPyMongoDBLSTM
- Trained an unsupervised LSTM autoencoder on 41,639 CNC servo-speed readings and evaluated 9,432 held-out test readings, flagging 48 anomalies whose MAE reconstruction error exceeded a 0.7056 threshold set from training loss.
- Ingested a 4GB CNC telemetry dataset into MongoDB through chunk-based Pandas processing without loading the full dataset into memory, enabling scalable querying and analysis of 112K+ servo-speed observations for downstream signal preprocessing.
- Built a time-series preprocessing pipeline with Pandas, NumPy, MongoDB, and linear interpolation, filling missing telemetry values and converting irregular signals into cycle-aligned sequences using CNC main-program execution intervals as consistent model inputs.