Animesh Giri

Technologies:

Key Highlights:

• Built ETL-style pipeline for multi-table retail demand data
• Engineered lag, rolling, calendar, and price features
• Trained XGBoost model with MAE, RMSE, and RMSSE validation
• Added safety stock, reorder point, EOQ, and Monte Carlo risk simulation
• Delivered API and dashboard interfaces

Technologies:

Key Highlights:

• Supports dataset validation, training, evaluation, and reporting workflows
• Includes configurable experiments for age prediction and model benchmarking
• Provides Streamlit UI for data, training, reports, benchmark results, and QA
• Designed for reproducible medical ML experimentation

Technologies:

Key Highlights:

• Demonstrates applied NLP and recommendation workflow
• Useful for showing semantic search and user-facing AI interaction design

Technologies:

Key Highlights:

• Demonstrates document intelligence, text processing, and model evaluation

Technologies:

Key Highlights:

• Event-driven architecture: FastAPI producer publishes to Redpanda, worker consumes and triages
• OpenAI worker generates priority, routing, summary, recommended action, and customer response
• Fallback logic handles API failures without dropping tickets
• Full stack Docker Compose deployment -- one command to run locally
• Live dashboard with filter by priority and GenAI provider

Technologies:

Key Highlights:

• Planner-solver-reviewer workflow with explicit handoff tags and shared context blackboard
• Provider abstraction for local, OpenAI, and Anthropic models
• Benchmark runner over JSONL task sets with exact, numeric, and keyword scoring
• FastAPI endpoint, Postgres persistence, and MLflow experiment tracking
• Local demo runs without API keys

Technologies:

Key Highlights:

• Compared Logistic Regression, XGBoost, Naive Bayes, and Neural Network on 5,110 patient records
• Applied SMOTE to training split only to prevent data leakage into test set
• Best model: Logistic Regression + SMOTE, ROC-AUC 0.706
• Showed accuracy is misleading on 4.2% positive-rate data -- all no-SMOTE models collapse to majority class

Technologies:

Key Highlights:

• XGBoost and logistic regression classifiers trained on limma differential expression genes
• Limma-based differential expression pipeline in R for LUAD vs LUSC contrast
• SHAP-ranked gene importance browser for model interpretability
• REST API with PostgreSQL backend over TCGA LUAD/LUSC cohort