Shreya Biswas

Built Python tool to screen 4,000+ IT records for quantitative model risk via 3-tier fuzzy matching (regex, Levenshtein, token-set) across 12 risk domains; flagged 399 candidates using 90th-percentile calibration, delivered as auditable Excel workbook. Validated Commodities IPV (Delta & Vega) via SABR calibration to reconcile FO vs. market valuations; flagged discrepancies for Market Risk Control. Tech Stack: Python, Pandas, NumPy, Excel, VBA, LaTeX, Bloomberg, Text Processing, Fuzzy Name matching

Built RAG pipeline over enterprise audit corpora using LangChain ParentDocumentRetriever (800/256-token hierarchical chunking), improving semantic recall by 34%. Hybrid dense (all-mpnet-base-v2) + BM25 EnsembleRetriever (70/30) boosted domain keyword coverage 58%→82%. RAGAS eval: 0.78 answer relevancy, 0.84 contextual precision, 3× faster lookup. Extracted sentiment topics via BERTopic with HDBSCAN hyperparameter tuning. Tech Stack: Python, LangChain, BERTopic, BERT, SpaCy, NLTK, NumPy, Pandas, Matplotlib

Designed event-driven trading pipeline using OAuth2/AMQP (RabbitMQ) to ingest real-time OTC data into Azure Data Lake via Function Apps. Applied ML-driven spread thresholds to route buy/sell signals, with Dead Letter Queue for zero-loss retry. Reconciled trade IDs/timestamps via Azure Data Factory, writing audit records to Azure SQL; surfaced P&L, counterparty credit, and latency alerts via Application Insights. Tech Stack: Python, NumPy, Pandas, Plotly, Azure Functions, ADF, ADLS, Azure SQL, RabbitMQ

Built XGBoost+RF PD model (80%+ recall, 5% imbalanced via SMOTE-TOMEK, OOT validated Jan'21–Nov'23); preprocessed multi-source credit data, reduced features 54→28 via SHAP/Gini; PSI monitoring + MLflow versioning. Detected 12 fraudulent customers in 12M records via Fuzzy Name Matching, raising 9 UARs with BFCR. Migrated SAS→Python on GCP (Airflow), built Looker dashboards, contributed 50K+ lines via Git. Tech Stack: Python, Scikit-learn, PySpark, MLflow, BigQuery, GCP, Airflow, Looker, FastAPI, Docker, SQL