Projects
- Built in-memory Redis-style server and CLI client in Go which supports redis core commands.
- Implemented thread-safe storage for strings (with expiry), lists (with blockings operations), sorted sets.
- Supports multiple concurrent clients and load an existing RDB file for persistence.
- Built a BitTorrent client that allows downloads using a single-file torrent or magnet links
- Implemented p2p protocol with tracker communction, piece checks, metadata exchange
- Built a git client that clones repositories using the git smart http protocol.
- Handles git objects and writes files directly, eliminating the need for ZIP extraction.
- Built a caching reverse proxy CLI tool which reduces server hits and request latency.
- Implemented a thread safe concurrent disk based cache storage system.
- Built a backend service that periodically checks website health and tracks uptime status over time.
- Set up background cron jobs to automatically schedule and run health checks every hour.
- Created REST APIs for managing monitored sites and viewing their current uptime status.
- Built a Slack integration to instantly alert users whenever a monitored site goes down.
- built a custom load balancer using round-robin scheduling to distribute traffic across servers.
- Added automatic health checks and reroute requests to healthy servers.
- bytehell is a UNIX-like command-line shell used to execute built-in Linux commands.
image-denoiser
- Designed an autoencoder-based image denoiser for low-light photos, reaching a PSNR of 28.256 dB.
- Applied RIDNet to improve visibility and detail retention, raising PSNR to 30.579 dB on the same set.
stock-sentiment
- Built an NLP-driven sentiment model to predict stock-price movements from financial news headlines.
- Back-tested the trading strategy on historical data, yielding 17.12% returns and a 1.34 Sharpe ratio.
Research Project: Estimation of Remaining Useful Life of Bearings using ML
Supervised by Prof. S.P. Harsha
- Built a predictive-maintenance framework estimating the Remaining Useful Life of high-speed bearings.
- Collected vibration signals from an experimental test rig and denoised them via wavelet thresholding.
- Extracted 34 multi-domain features and filtered them using S-weighted and Pearson correlation scoring.
- Designed a self-attention-augmented BiLSTM to model temporal degradation patterns in vibration data.
- Tuned hyperparameters via Bayesian optimisation, selecting hidden units, dropout, and learning rate.
- Achieved R² of 0.825 on combined data and 0.781 on the held-out bearing under leave-one-out testing.
Research Project: Simulation of Human-Induced Load Patterns using ML
Supervised by Prof. Anil Kumar
- Developed a GAN-based model generating human-induced load patterns at 98% similarity to real data.
- Processed 658+ experimental samples across walking, jumping, and bouncing loads at 1.5 to 3.5 Hz.
- Built a Forecasting+LSTM hybrid model and studied ARIMA and Prophet for time-series load synthesis.
- Validated synthetic loads via PSD replication and statistical equivalence in structural response.