qjian33

Live demo: drg-hospital-demo.streamlit.ap｜ Code: github.com/qjian33/drg-hospital-demo Overview: A two-level interactive dashboard (region overview + single-hospital drill-down) on 500K+ annual DRG settlement records, covering 22 core operating metrics (CMI, cost per case, cost per weight, surplus rate, fund compensation ratio, time/cost consumption indices) with year-over-year comparison. Key Responsibilities & Outcomes: DRG Analytics System: RW-tier P&L, discipline-development quadrant (advantage/potential/weak/problem groups), TOPSIS composite department ranking, mortality risk-group monitoring, and a 6-rule case-quality QC; original 'false-profit' detection (total-cost overrun but positive fund surplus → self-pay cost-shifting alert). De-identification & Front-End Engineering: Full de-identification (hospital/patient/time); self-built browser-side aggregation engine — a Web Worker streams and aggregates 620K rows without freezing the UI, building dashboards instantly from CSV/XLSX uploads. Result: Browser-side aggregation verified 0-error vs the Python pipeline across 18 metrics; single-file, self-contained, runs offline.

Live demo: onchain-defi-intel.streamlit.app ｜ Code: github.com/qjian33/onchain-defi-dashboard Overview: A seven-module DeFi analytics dashboard on live public data (450+ chains, 7,800+ protocols via DefiLlama/CoinGecko), each module auto-generating a plain-English verdict. Key Responsibilities & Outcomes: Analytics Modules: market structure, per-chain drill-down, cross-chain quant comparison (correlation / volatility / HHI concentration / capital rotation), whale-flow monitoring, and stablecoin de-peg alerts. Statistical Rigor: scale-invariant return-based anomaly detection (rolling ±kσ), OLS trend extrapolation with 95% prediction intervals, TVL flow-vs-price decomposition; unit-tested risk-metric suite (Sharpe/Sortino/VaR/CVaR).

Overview: built a pharma decision tool across Market Access / Medical Affairs / Commercial to answer 'whom to sell to, on what evidence, and at which hospitls.' Key Responsibilities & Outcomes: Market Sizing: Drill-down from therapeutic area to ICD 3-digit sub-indications; introduced a 'primary-diagnosis vs all-inclusive' reconciliation that reconstructs the true chronic-disease market — counting diabetes/hypertension as comorbidities expands endocrine-related admissions from 32K to 200K cases and related drug spend from CNY 28M to CNY 415M. Golden Targets: Dual thresholds of high loss rate (>40%) + high drug ratio (>25%) isolate 69 disease groups — the easiest access entry points for pharmacoeconomically-advantaged products. Hospital Targeting & Evidence: A 'cost-consumption index' (a hospital's per-case cost divided by the all-hospital average) benchmarks priority hospitals into a KA battle map; complemented by comorbidity networks, readmission rates, clinical outcomes, and patient out-of-pocket burden. Result: Delivered an interactive dashboard + market-insight report + metric-definition documentation; fully de-identified with transparent, auditable metrics.

Overview: Developed an intelligent crypto investment-research tool that converts quantitative feature data into professional research reports via LLMs, achieving a fully automated cross-platform publishing pipeline with zero manual intervention. Key Responsibilities & Outcomes: LLM Analysis & Dialogue Engine: Integrated the DeepSeek LLM to process structured quantitative data; used advanced Prompt Engineering to auto-parse order flow and funding-rate anomalies and extract trading signals; built a topic-dialogue channel with an reliable LLM-powered text-generation pipeline to suppress hallucinations and produce high-quality research insights. Multimodal Chart-Text Processing: Used Python backend scripts to structurally assemble high-frequency quantitative monitoring charts with LLM-generated strategy commentary, auto-generating dense analytical summary cards. Automated Distribution Engine: Built an autonomous distribution controller using scheduled tasks to seamlessly push graphical reports to Telegram channels and Binance Square, fully replacing manual posting and significantly improving multi-platform, multi-account operational efficiency.

Background: The hospital was running at a significant loss and required a comprehensive operational analysis to identify root causes. Key Responsibilities & Outcomes: Multi-Dimensional Drill-Down Analysis: Decomposed performance across disease types, case categories, departments, cost structures, and DRG weight tiers; pinpointed root causes including poor cost control in internal medicine (excess drug/exam spend) and significant losses in ICU/pediatrics; proposed targeted strategies such as refined pediatric groupings and ICU subsidy increases. Dynamic Early-Warning Dashboard: Rapidly built a real-time Metabase monitoring dashboard delivering modules including 'Fund Operations Overview' and 'Drug Cost Ratio Tracker'; implemented automated overspend alerts, improving risk detection timeliness to T+1. Result: Recommendations were adopted by hospital leadership; the hospital returned to profitability in October of that year, achieving a surplus of over CNY 3 million.

Overview: Core responsibilities included DRG grouping simulation, data quality control, and payment settlement modeling for medical insurance policy evaluation. Key Contributions: Established a standardized data quality-control pipeline using SQL for cleansing and extraction; simulated medical insurance payment outcomes under varying policy rules in R to quantify policy impact and support the bureau in setting payment weights and benchmarks. Result: Delivered robust data evidence for policy decisions, contributing to a 6% reduction in average patient medical costs.

Reporting Automation: Architected a Metabase-based automated hospital operations dashboard with real-time subscriptions and anomaly alerts, compressing the data feedback cycle from monthly to T+1 and reducing routine reporting workload by 40% across departments. Decision Automation: Independently built an R Shiny web application for automated medical-insurance operations analysis, enabling on-demand diagnostic reports in seconds. Routine Operations Monitoring & Policy Simulation: Performed deep statistical modeling of medical-fund performance across multiple cities; used SQL for large-scale data cleansing, clinical clustering, weight calculation, and risk profiling; supported health bureaus in building disease catalogs and policy evaluation reports. Healthcare Analytics: Monitored operational, performance, and P&L metrics for medical institutions; identified root issues through trend decomposition and benchmarking; delivered strategic recommendations via professional analytical reports.

Data Source Evaluation: Assessed quality, availability, and accuracy of third-party data sources; collaborated with the team using R and Excel to optimize risk engineering models. Risk Monitoring: Built binning analysis and delinquency-rate reporting frameworks; extracted anomalous risk data via SQL and diagnosed core contributing factors.

Overview: Applied sentiment mining and churn modeling to user feedback text to optimize product strategy and improve customer satisfaction. Key Responsibilities & Outcomes: NLP Text Mining: Cleaned social network text with Python; applied text classification and semantic analysis to build a negative-review knowledge graph; pinpointed key pain points including ambiguous inter-party service responsibilities and difficulty scheduling technician appointments. Churn Prediction Modeling: Conducted EDA and built a logistic regression classifier; demonstrated that price, geography, and plan type were the most significant churn drivers. Result: Collaborated with the business team to redesign customer service response paths and clarify responsibility boundaries, improving customer satisfaction scores by 10%.