healthcare operations delivery
Shailesh Dudala
Production-grade AI systems across regulated workflows, local models, forecasting, and MLOps.
Senior Applied AI / ML Engineer
I turn professional delivery, hackathon wins, local model experiments, and research prototypes into production-aware AI systems across healthcare, insurance, document intelligence, forecasting, agentic workflows, and MLOps.
Healthcare + insurance | hackathon wins | local LLM labs | LangGraph/RAG/OCR | FastAPI/Azure/MLflow | XGBoost/backtesting
Agentic AILLM steering workbench
fhir_bundle.jsonHealthcare AISynthetic care episode
"resourceType": "Bundle","encounters_180d": 3,"discharge": "SNF"claims_document.pdfClaims automationDocument to data workflowclaim_type: medical_reviewconfidence_gate: enabledroute: analyst_queueagent_trace.yamlAgentic AITool workflow stateroute: extract -> verifytool_calls: 6citation_check: passedforecast_backtest.jsonForecastingCalibration and model comparison"games": "1999-2024","baseline": "elo","metric": "brier_score"model_card.mdResponsible AIGovernance boundarySynthetic demo onlyNo PHI or PIIHuman review requiredpytest_results.logMLOpsBuild and health evidencepytest: 42 passeddocker: build okhealth: status=okContractsFHIR, API, model, and response schemasWorkflowsextraction, agents, review, and evaluationDeliveryDocker, tests, health checks, and public demos
Selected outcomes
Delivery metrics from healthcare, insurance, and applied AI systems work.
Selected outcomes from healthcare, insurance, hackathon builds, local model labs, and applied AI delivery. Public repositories are synthetic, sanitized, historical, or research/demo implementations only.
Operating proofMetrics first, then the systems behind them.
Backlog clearance, review acceleration, waste reduction, and deployment speed set the context for the case-study flow below.
claims and document review acceleration
automated closure improvement
fraud, waste, and abuse analytics
quality and value-based care programs
MLOps and delivery modernization
Domains
A connected map of the systems I build across regulated AI, document intelligence, forecasting, local model labs, and MLOps.
Each row links a domain to a concrete system, the implementation stack behind it, and the next case study a reviewer can inspect.
010203040506
Healthcare AI
Healthcare & Interoperability AI
FHIR R4, HL7 event flows, readmission risk, LOS, ED utilization, care gaps, quality programs, and reviewable risk outputs.
FHIR R4HL7 ADT/ORU/MDMHEDISFastAPI
Example systemHospital Readmission FHIR ML API
FHIR / HL7 eventFeature contractRisk APIReviewable response
Synthetic FHIR input to typed risk response
Claims automation
Claims & Document Automation
OCR, structured extraction, validation gates, confidence scoring, exception routing, and document-to-data workflows.
OCRTesseractLangChainReview queues
Example systemFreshTrack AI + sanitized claims patterns
Image / PDFOCR cleanupValidation gatesAnalyst queue
Noisy document inputs routed into structured review
Agentic AI
Agentic AI & LLMOps
LangGraph-style orchestration, tool calling, local LLMs, activation steering, evaluation gates, and observability surfaces.
LangGraphOllamaPyTorchReact
Example systemLLM Steering Lab
Prompt pairTool routeEval gateWorkbench trace
Workbench traces for steering and evaluation
Forecasting
Quant, Forecasting & Backtesting
AlphaQuant, sports forecasting, Elo, XGBoost, calibration, model comparison, Brier score, and historical backtesting.
XGBoostEloBrier scoreBacktests
Example systemAlphaQuant + Sports Forecasting Lab
Signal historyBaselinesModel comparisonBacktest report
Historical model comparison with calibration notes
MLOps
MLOps & Infrastructure
Dockerized APIs, CI checks, MLflow-style delivery, Azure patterns, Kubernetes readiness, health checks, and deployment discipline.
DockerCI/CDAzureMLflow
Example systemSystems Library
API contractContainer buildCI checkHealth surface
Deployable patterns, tests, and health evidence
Research foundations
Research & Data Science Foundations
Biomedical informatics, visualization, GPU planning, scientific ML, time-series analysis, and earlier experimental systems.
scikit-learnVisualizationGPU planningRAG
Example systemResearch Archive
Research questionDataset shapeModel or viewReusable note
Foundational experiments organized by technical theme
Featured work
Case-study systems that turn the domain map into visible architecture and working evidence.
A flagship healthcare ML API anchors the page, then agentic AI, quant research, interoperability, document AI, and local RAG show the breadth around it.
Flagship case study
Hospital Readmission FHIR ML API
Synthetic FHIR-style scoring service for 30-day readmission risk with schema validation, feature extraction, explainability, tests, model-card framing, and Dockerized FastAPI deployment.
FastAPIPythonFHIR-style JSONXGBoost patternpytestDocker
FHIR ML servicetyped risk contract
LLMOps / interpretability / local-first AI
LLM Steering Lab
LLM control work can become vague when reduced to prompt examples. This lab treats behavior steering as reproducible experiments, model registries, hook points, UI controls, and documented limitations.
Open case study
Agentic market intelligence / quant research
Agentic Alpha Engine
Research workflows need ingestion boundaries, storage, retrieval, planning, verification, state, and report contracts before any conclusion is trusted.
Open case study
Sports forecasting / model calibration
NFL Forecasting Lab
Forecasting systems are credible only when baselines, model comparisons, calibration, and historical backtests are visible.
Healthcare interoperability / quality intelligence
HL7 AI Challenge
Clinical quality and care-gap workflows often depend on fragmented events and manual chasing. This challenge platform shows how HL7-style events can become FHIR-aligned resources and downstream intelligence.
Computer vision / OCR / structured extraction
FreshTrack AI
Receipt and image parsing pipelines need to distinguish real entities from payment lines, addresses, totals, headers, and formatting noise.
Document AI / local LLM extraction
Local Document AI
Regulated document workflows need extraction assistance without automatically sending sensitive files to remote services.
Evidence
Evidence organized as a guided dossier, not a pile of disconnected cards.
Move through UI demos, JSON contracts, charts, diagrams, and local-model workflows in sequence, then open the larger preview when a reviewer needs the artifact detail.
AlphaQuant UI Demo
Local-first market intelligence workbench with planner state, storage fabric, verification, and structured reports.
Professional systems
Enterprise patterns behind the public systems.
Sanitized architecture streams connect the portfolio to claims automation, HEDIS evidence extraction, local RAG/OCR, predictive models, FWA analytics, and healthcare analytics platforms.
Enterprise bridge
Public case studies point to the same architecture habits used in regulated delivery.
The systems below translate professional healthcare and insurance work into sanitized patterns: intake, extraction, validation, review, analytics, deployment, and accountable handoff.
7K backlog cleared90% review-time reduction18% FWA waste reduction$3M P4P impact
Enterprise systems are described as sanitized architecture patterns. No PHI, PII, employer-confidential data, or proprietary workflows are included.
010203040506
Enterprise system
Claims Document Automation / Autonomous Adjudication Support
LangGraph-style orchestration for document understanding, structured extraction, validation gates, exception routing, and downstream claims-system handoffs.
document intakeOCR/extractionagent graphvalidation gatesexception routing
LangGraph patternsAzureADLSCosmos DB
Claims review acceleration and safer automation boundaries without exposing employer data.Enterprise system
HEDIS Evidence Extraction Pipeline
Quality-measure evidence workflow over HL7 MDM/ORU-style documents, base64 PDFs, OCR, rule validation, human review, and closure tracking.
HL7 MDM/ORUbase64 PDFOCRmeasure rulesevidence extraction
PyMuPDFTesseractFHIR/HL7rules engine
Automated closure improvement while preserving reviewer accountability.Enterprise system
On-Prem RAG/OCR Compliance Review Microservice
Private inference workflow for regulated document review using local SLMs, retrieval, OCR, containerized services, and analyst-facing review surfaces.
local document storeOCRchunkingretrievalGemma/Ollama/vLLM
PodmanStreamlitOllamavLLM
Review-time reduction and backlog clearance without moving private documents into public demos.Enterprise system
Readmission / LOS / ED Utilization Predictive Models
Healthcare predictive modeling across claims, ADT events, demographics, diagnoses, medications, SDoH, risk scores, and operational features.
claims + ADTfeature storeXGBoost/stat modelsrisk outputsPower BI
PythonSQLXGBoostscikit-learn
Care-management prioritization, utilization insight, and governed model delivery patterns.Enterprise system
FWA Transportation Anomaly Detection
Fraud, waste, and abuse analytics over multi-leg trips, utilization anomalies, geospatial validation, and proactive review workflows.
trip eventsmulti-leg featuresESRI validationanomaly scoringreview queue
PythonSQLESRIanomaly features
Waste reduction through targeted, explainable review signals.Enterprise system
Healthcare Analytics Platform / Health Index
0-to-1 healthcare analytics platform work using composite risk scoring, care gaps, SDoH features, quality analytics, and embedded delivery.
member datarisk scoringcare gapshealth indexembedded dashboard
PythonSQLPower BI EmbeddedAzure
New revenue and client impact through operational healthcare analytics products.Implementation details
Follow the build trail from case studies to source repos and enterprise patterns.
Use the detailed pages to inspect architecture, artifacts, responsible-use boundaries, and the systems library behind the landing page.