Collaborative Developer Productivity Workspace

Real-time MERN workspace with collaborative coding sandbox, AI-assisted workflows, and high-performance search.

• Socket.IO collaborative editor — presence indicators, typing awareness
• Gemini API for AI-assisted code explanation and refactoring
• MongoDB Atlas Search — sub-10ms latency, ~40% query improvement
• Docker deployment, JWT auth, session-scoped activity logs

ML-Powered Event Discovery, Recommendation & Booking

Full-stack platform aggregating tech events and college societies with personalised ML recommendations.

• Hybrid ML microservice — TF-IDF + collaborative filtering, cold-start handling
• Race-condition-free booking engine — MongoDB atomic ops, 100% consistency
• NextAuth v5 OAuth, optimistic UI updates, dynamic OpenGraph images
• Role-aware access control across organiser and attendee flows

C-Subset to Assembly Compiler

Multi-stage compiler for a C subset written in C++ — parser, semantic analysis, IR, and optimizations.

• Hand-written recursive-descent parser, AST, type checking, scope resolution
• 3-address IR (TAC) with basic-block CFG construction
• Constant folding, dead-code elimination, copy propagation — ~25% instruction reduction
• 100+ pytest cases covering pointer arithmetic, nested scopes, register spilling

ANN with TensorFlow & Keras

Neural network model predicting customer churn on 10,000 bank records with interpretability analysis.

• 3-layer ANN (64→32→16) — 86.2% accuracy, AUC-ROC 0.87
• Dropout, BatchNorm, L2 reg, EarlyStopping, ReduceLROnPlateau
• Threshold optimization, permutation importance, Adam vs SGD vs RMSprop
• Projected $4M+ annual retention saving from churn reduction

Time Series Classifier — Research Implementation

Implementation and analysis of Proximity Forest (Lucas et al., 2019) — benchmarked on 85 UCR Archive datasets.

• Gini-impurity splitting with stochastic selection from 11 elastic distance measures per-node
• Eliminated O(n²) cross-validation overhead of Elastic Ensemble baseline
• Confirmed 10⁵× speedup on 1M-series SITS dataset — O(n log n) vs O(n²·l³)
• 32-page technical report with complexity visualisations and critical-difference diagrams

IIT Ropar Ed-Tech Platform Contribution

Open-source contribution to Vibe — an IIT Ropar platform for adaptive student assessment.

• Replaced brittle sys.exit(1) paths with structured PipelineStep abstraction
• Interactive Retry/Skip/Abort UI with persistent JSON state tracking
• Preflight port validation across 6 critical services — eliminated silent setup failures
• Unified stdout/stderr logging and --clean CLI flag for deterministic env reset