AI Researcher and Full-Stack Software Developer

Designed an end-to-end behavioral modeling pipeline combining Big Five personality data, experience sampling methods, LSTM models, Temporal Fusion Transformers, and SHAP explainability for cognitive-pattern research.

Built MC Dropout uncertainty quantification and a cognitive translation layer that converted feature attributions into interpretable psychological predictions. Processed more than 1,000 data points daily and reduced prediction error by 23%.

Designed a clinical validation framework for ADHD and burnout digital phenotyping, achieving 92% user comprehension through SHAP-based explanations and preparing the work for open-science publication.

Built an AI research intelligence and synthesis engine using Node.js, FastAPI, Next.js, Docker, RAG workflows, and Gemini/LLM integrations. The system ingested more than 200 scientific papers, extracted metadata, and identified research gaps, novelty scores, and feasibility-scored hypotheses.

Implemented contextual question answering, semantic search, and pattern discovery workflows to help transform broad topics into targeted research questions. Exposed more than 25 REST APIs for document ingestion and optimized the backend with Redis caching, reducing latency by 45%.

Containerized the platform with Docker to support horizontal scalability.

Conducted a medical imaging study using transfer learning with ResNet-50 and PyTorch across more than 3,000 MRI scans for brain tumor detection. Applied Grad-CAM explanations that achieved 89% alignment with radiologist annotations and MC Dropout uncertainty estimation that flagged ambiguous cases for expert review.

Fine-tuned BERT for IMDB sentiment analysis on 50,000 reviews, achieving 95% accuracy and 94% F1 score. Applied LIME explanations with 87% human agreement to make NLP model predictions understandable for non-expert stakeholders.

Developed an event-driven multi-asset algorithmic trading backtester with market microstructure signals, momentum regimes, position management, order sizing, risk limits, and transaction-cost modeling.

Applied XGBoost, LightGBM, and SHAP for market-predictive feature discovery using lagged returns, volatility, and cross-asset correlations. Used walk-forward validation and time-aligned cross-validation to reduce lookahead bias and overfitting.

Built a real-time financial market data and prediction platform integrating Indian and US equities, cryptocurrency feeds, derivatives pricing, news sentiment, and technical indicators. Trading strategies generated $9,900 in cumulative competition returns.

Built SwiftTrack, a real-time logistics platform using Next.js, TypeScript, Supabase, PostgreSQL, Socket.IO, and WebSockets. Supported live GPS tracking for more than 100 concurrent users, reduced bandwidth by 25%, and improved query performance by 40% through schema indexing.

Developed GearLink CRM with Next.js, Firebase, Google Gemini, and Genkit, including POS, inventory, invoices, Kanban sales pipelines, automated reporting, AI-powered part recommendations, and marketing email generation.

Created Expenseasy with Next.js, PostgreSQL, Prisma, Tesseract.js, and Vercel, implementing role-based expense workflows, OCR receipt parsing, real-time currency conversion, and a production-ready database backend. Also built Lakshya360, an offline-capable PWA career counselor chatbot that was a national hackathon finalist.

B.Tech in Computer Science Engineering with a minor in Robotics and Automation.