Hi, i'm parv

Advanced Engineering Hub Intern · Bengaluru

• Contributed to the development of an enterprise Identity & Access Management (IAM) solution by implementing explainable machine learning pipelines for identity risk scoring and access governance.
• Engineered risk-scoring features and benchmarked ensemble models (XGBoost, LightGBM) to improve identity risk classification accuracy, applying feature engineering and model explainability techniques from prior transformer-based forecasting work.

Project Intern — Hesor, PPG Vital-Sign Monitoring · Palakkad, Kerala

• Lifted the blood-pressure classification path out of a PyQt5 research app into a standalone FastAPI service, loading the model once at startup and validating each request with Pydantic so malformed or mismatched IR/RED blocks are rejected before they reach inference.
• Rebuilt the inference stage to resample every 5-second block from 500 Hz to 125 Hz, run Chebyshev filtering and peak-onset feature extraction, and return a three-class result (Hypotension / Normotension / Hypertension) with a finger-detection gate that skips blocks instead of guessing.
• Wrote a parity harness that replays identical inputs through both the new service and the original research implementation, so the migration could be proven numerically equivalent rather than assumed correct.
• Built the Expo/React Native client that streams PPG at 500 Hz from an HC-05 sensor over Bluetooth Classic SPP through a custom Kotlin RFCOMM native module, buffering samples in refs and flushing to React state at 10 Hz to hold 60 FPS on the live waveform charts.
• Implemented the TypeScript signal-processing layer end to end — DC removal, bandpass filtering, AMDF peak detection, and the derived metrics (pulse rate, SpO₂, perfusion index, signal quality, rhythm classification, and HRV: MeanPP, SDPP, RMSSD, pPP50).
• Extended the dashboard with longitudinal trend charts and environmental context (weather and AQI) so readings can be interpreted against the conditions they were taken in.