Stop wasting your AI coding CLI usage windows. Queue tasks for spare capacity + prewarm your 5-hour window so resets land when you need them. Claude Code, Codex CLI, Cursor.

Flight recorder and firewall for AI agents — records every tool call, tracks lethal-trifecta taint across a session, and blocks the prompt-injection exfiltration path. Zero deps, cross-harness.

Enterprise healthcare data pipeline: real-time streaming (Kafka+Spark) + batch ETL processing 10M+ patient records. Airflow orchestration, dbt transformations, Great Expectations validation. SCD Type 2, Terraform IaC (Snowflake/GCP), Prometheus/Grafana monitoring. Production-grade with full testing & documentation.

High-performance skin cancer classification using EfficientNet-B3 achieving 91% accuracy across 35 conditions on 245K images. Implemented multi-GPU parallelism with 4× NVIDIA A100s for 3.3× training speedup and 84% parallel efficiency optimization.

NBA player injury prediction system with 78% AUC-ROC. Complete MLOps pipeline: 4.5K+ games processed with 55+ engineered features, FastAPI serving <100ms predictions, Redis caching, MLflow experiment tracking, Prometheus/Grafana monitoring. Production-ready with Docker/K8s deployment, automated testing (90%+ coverage), and real-time API endpoints.

Production-grade bank reconciliation agent with invariants, retries, traces, replays, and evals

Quantitative trading simulator with smart order routing (SOR) and execution algorithms (VWAP, TWAP, POV). Features modular Streamlit backtesting framework analyzing execution costs and market microstructure with real-time Yahoo Finance data integration.

Interactive supply chain dashboard using Neo4j knowledge graphs and Streamlit for real-time analysis. Features natural language querying with LLMs translating to Cypher queries, enabling conversational analytics for supplier traceability and shipment management.

Implemented a Deep Q-Network (DQN) based reinforcement learning system for automated pairs trading. Built a custom Gym-like environment with real-time feature engineering on stock spreads, volatility, and rolling means, enabling dynamic decision-making and optimized portfolio performance in varying market conditions.

AI-assisted video edit detector — NLP + FAISS RAG CLI mirroring Descript Underlord

Built a CNN-based image forgery detection system using CASIA 2.0 dataset, achieving 96.9% accuracy. Deployed model on AWS EC2 for accessibility, with SVM for classification and a user-friendly GUI web app that allows users to test images for copy-move manipulation detection in real time.

No description.

My personal repository

A high-throughput and memory-efficient inference and serving engine for LLMs

Semantic deduplication POC for DatologyAI Solutions Engineer assignment - CLIP embeddings for e-commerce product catalog curation

snowconvert-ai-challenge

Fireworks Assignment: Vinay Jogani

Describe your task, get one AI model recommendation, backed by benchmarks and real developer sentiment.

Qode Assessment 2

Qode_Assesment

A/B testing platform with frequentist (t-test, chi-square), Bayesian, and sequential testing (O'Brien-Fleming). ML models: Random Forest for CLV prediction and churn classification. SQL analytics: window functions, CTEs, cohort retention, RFM segmentation. PostgreSQL (100K users, 1M bets). Stack: Python, scikit-learn, pandas, statsmodels, scipy.

A machine learning-powered sports analytics platform that optimizes betting strategies using the Kelly Criterion across multiple sports. Integrates real-time odds data with custom ML models to provide data-driven bankroll management and predictive insights, achieving 87% prediction accuracy

A powerful framework for building realtime voice AI agents 🤖🎙️📹

Recipes and resources for building, deploying, and fine-tuning generative AI with Fireworks.

No description.

Full stack financial analytics system: PostgreSQL database with normalized tables processing 1,750+ transactions. Built 12 advanced SQL queries (CTEs, window functions, recursive queries). 5-page interactive Tableau dashboard. Python ETL pipeline with pandas. Automated reporting. Demonstrates database design, statistical analysis & BI development.

Production-ready microservices API for personal finance management with expense tracking, budgets, and analytics. Built with FastAPI, PostgreSQL, and Redis. Features 4 independent services, 90%+ test coverage, CI/CD with GitHub Actions, Docker/Kubernetes deployment, rate limiting, and Redis caching. Demonstrates enterprise software engineering.

AI-powered code explanation system using fine-tuned CodeT5 with Retrieval-Augmented Generation. Generates natural language explanations for Python code with 24% BLEU improvement over GPT-4, interactive Q&A capabilities, and sub-500ms response times on 2M+ training examples.

Smart trading assistant powered by Roboquant and Streamlit. Analyze stocks with 4 technical indicators (RSI, MACD, SMA, Bollinger Bands), get buy/sell signals, price targets, and risk management suggestions. Real-time data from Yahoo Finance with interactive charts.

Developed an AI-powered system combining a biography knowledge assistant with an intelligent pairs trading bot. The solution parses unstructured text for contextual Q&A and applies machine learning with statistical arbitrage for automated trade decision-making

Domain-specific chatbot parsing Swimlane diagrams and OpenAPI specs into structured event sequences. Features transformer-based semantic retrieval with FAISS vector store and modular production pipeline for enterprise workflow automation.

No description.

Volatility surface modeling with Black-Scholes and Heston calibration using SciPy/CVXPY optimization. Built arbitrage-detecting trading strategies with improved Sharpe ratios, comprehensive VaR/ES risk analysis, and real-time Bloomberg/yFinance data integration.

Custom Gymnasium RL environment for portfolio management across 5 synthetic assets with realistic trading costs and volatility. PPO agent trained with Stable-Baselines3 achieved 23% return improvement over baselines while maintaining 90% diversification consistency.

Advanced options pricing models implementing Black-Scholes, Binomial Tree, and Monte Carlo methods. Features real-time data analysis via yfinance API, multi-method comparison tools, and optimized computation with NumPy/SciPy for scalable financial modeling.

Deep reinforcement learning trading algorithm using OpenAI Gym environment with financial time-series data and technical indicators. Built with stable_baselines3, featuring Bayesian hyperparameter optimization for high-frequency trading decisions and performance visualization on historical market data.

Developed a pipeline to detect potential insider trading by integrating SEC Form 4 filings with stock market data via yfinance. Applied ETL processes, engineered financial features (e.g., moving averages, PVT, buy-sell ratios), and trained ML models (Random Forest, SVM) to flag anomalous trading patterns.

Built a DeepFake detection system using GAN-based models to classify manipulated content via perceptual cues. Achieved 91% accuracy with a plain-frames model, outperforming MRI-GAN SSIM at 74%. Explored loss functions, hyper-parameter tuning, and perceptual metrics to improve robustness against synthetic media.

Developed an ESG-integrated mean-variance optimizer using PyPortfolioOpt on S&P 500 stocks. Applied covariance shrinkage and CAPM-based returns to generate efficient frontiers under ESG thresholds, reducing portfolio volatility by 22% and improving Sharpe ratio by 18%, proving the value of sustainable investing.

Developed an end-to-end stock analysis and recommendation system combining technical indicators with sentiment analysis from Reddit and Twitter. Automated data pipeline integrated Yahoo Finance API, NASDAQ data via FTP, and social media insights to rank top 5 daily stock picks.

A collection of applied data science projects covering causal inference, exploratory data analysis, and visualization. Includes studies on air pollution impact, product recommendations, FIFA player analytics, Titanic survival insights, and tennis performance metrics.