Anurag Singh

Site Curation

• Developed a US site recommendation model utilizing claims data and CTGov site comparison.
• Integrated real-time outcome visualization service to enhance site selection for clinical trials.
• Recommended sites based on patient demographics, tumor types, household roster, and other relevant factors.

Risk-Based Monitoring - Recommendation Engine

• Developed a predictive analytics pipeline to proactively detect clinical trial risks by analyzing historical KRIs and action-item commentary.
• Utilized advanced NLP techniques, including Sentence Transformers, KMeans clustering, and fuzzy matching, to classify and structure free-text data for enhanced modeling.
• Created temporal features through window-based aggregation and merge asof joins, significantly improving model accuracy and predictive performance.
• Built and optimized an XGBoost classification model (optimized for PR-AUC) to accurately predict protocol deviations, enabling early and informed risk interventions.
• Generated actionable insights using SHAP explain ability methods and designed interactive dashboards to facilitate strategic decision-making in clinical monitoring.
• Integrated LLM-generated recommendations by leveraging action-item topics and predictive outcomes to automate 'Next Best Action' guidance for intermediate-risk scenarios.

Proactive Site Management Tool I Healthcare Company

• Led a team of 6 data scientists to develop an AI-powered platform for managing clinical site communications, including interactions with HCPs, reducing preparation time for site visits by 60%.
• Designed and implemented a LangChain-based architecture integrated with GPT-40 to transform unstructured clinical site communications (with remote site monitors, HCPs, etc.) into actionable insights, achieving 85% accuracy.
• Collaborated with stakeholders to align technical solutions with business requirements, successfully transitioning from PoC to full-scale implementation with 100% on-time delivery.

Clinical Trial Landscape Tool I Fortune 100 Pharmaceutical

• Developed an advanced NLP pipeline using FlashText, reducing manual lookups of measuring scales by 70% and saving 25+ analyst hours weekly.
• Built and fine-tuned a custom Phi-2 model for automated endpoint extraction, eliminating 20+ hours of manual file creation weekly.
• Created a self-improving system where extracted data is tagged using FlashText and incorporated into training, enabling continuous performance gains with minimal oversight.

Physician Referral Network I Fortune 100 Pharmaceutical

• Built a Python pipeline to identify and network with Healthcare Providers (HCPs) across target markets, enhancing engagement and influence nationwide.
• Integrated claims data with NPPES web scraping and geocoding to improve HCP identification and engagement by 75%.
• Extracted and analyzed PubMed citation data to enrich HCP selection for rare disease expertise, increasing suitable candidate identification by 40%.

GBDS Dashboard IFortune 100 Pharmaceutical

• Developed Python scripts integrated with Tableau for resource allocation and budget tracking, streamlining data processing by 35%.
• Consolidated multi-source datasets using Python, SQL, and regex to deliver consistent, actionable insights across departments.
• Enabled leadership to monitor spending versus planned budgets across assets, improving financial oversight and resource allocation.

QC Dashboard I Fortune 100 Pharmaceutical

• Created an Excel-based dashboard with embedded business rules for data validation during database migration, reducing validation errors by 65%.
• Automated resolution of data discrepancies using macros and standardized business rules, ensuring seamless data refreshes.
• Maintained data integrity across systems, supporting accurate pharmaceutical records during critical transition periods.

Horizon Scanning Tool IFortune 100 Pharmaceutical

• Engineered a Python-powered web scraping tool to extract competing clinical trial data from ClinicalTrials.gov, enhancing market analysis capabilities.
• Automated real-time market data collection, reducing manual research effort by 50% and improving data currency.
• Delivered competitive landscape insights to inform strategic decision-making and research prioritization.