One internship, applied across the full ML lifecycle — not just modeling.
Six builds spanning agentic systems, explainable ML, and multilingual NLP — chosen as the strongest, most representative work across every iteration of this portfolio.
An LLM-powered agentic platform for multi-step task execution, built on structured prompt pipelines and modular API orchestration. A YAML-driven configuration layer lets new workflows be composed without touching code, and the whole platform is containerized with Docker for consistent deployment.
An end-to-end recommendation pipeline that doesn't just predict — it explains. SHAP interpretability surfaces why each career suggestion was made, served through a real-time FastAPI backend with feature engineering and ranking logic, and a Streamlit front-end for fast user testing.
Resume-to-job semantic matching using BERT embeddings combined with TF-IDF features and cosine similarity ranking. Extracts structured skills from resumes and runs a gap-analysis module to recommend learning paths — delivered as a Streamlit prototype for rapid iteration.
An intent-classification and entity-recognition pipeline built on multilingual transformer models, supporting 3+ languages with a translation layer for fallback cases. Packaged with Docker for consistent testing and evaluation, and exposed via REST API for channel-agnostic deployment.
Aggregated job-market signals from multiple portals to compute trending skills and role growth trajectories. Trained models mapping user skill profiles to probable transition pathways, visualized through interactive Tableau dashboards.
Tree-based ensemble models (XGBoost) for churn prediction, validated with cross-validation and AUC metrics. SHAP attribution surfaces the key drivers behind each prediction for business-facing retention planning.
Happy to walk through architecture decisions, trade-offs, or live-demo any of the above.