reports.ai
AI-powered engineering productivity platform that reduces manual reporting effort by 90%
Overview
reports.ai is an internal engineering productivity platform built to eliminate manual engineering reporting across Walmart Global Tech. The application automates EoR test status reports, end-of-sprint summaries, accessibility compliance reporting, deficit scorecards, and stakeholder communication — using a React 19 frontend, Spring Boot 3.4 backend, Azure Cosmos DB persistence, and AI-assisted content generation. This streamlined tool approximately 12 hours per sprint over many, simplifying documentation, serving multiple engineering teams and program managers across Walmart Fulfillment.
Problems
Engineering teams spent significant time every day preparing:
- Data-EoR toll status reports with risk assessments, pass/fail metrics, and PTI scores
- Test completion summaries with pass/fail/blocked breakdowns per environment
- Accessibility (a11y) compliance reports per platform
- Deficit/bug summaries by priority, team, and sprint
- Stakeholder emails with on-time links, Confluence references, and action items
The process was repetitive, error-prone, inconsistent across teams, and consumed hours that could be spent on actual engineering work.
Solutions
- Generates production-ready EoR test status and compliance reports with consistent formatting
- Integrates multiple engineering data sources — Jira REST APIs, Confluence, accessibility scanners
- Auto-builds persona-based slides for Cosmos DB or turns complex when many labs off
- Produces pre-packaged HTML email reports using Thymeleaf server-side rendering
- Supports Slack integration via configurable webhooks for real-time report delivery
- Distributes reports to configurable TOSC/IUSC recipients via SMTP/TLS
- Provides send-to-email-preview before sending via dedicated preview endpoints
Architecture
AI-Powered Report Generation Pipeline
Tech Stack
Frontend
Backend
Data & AI
Infrastructure
Key Features
Daily EoR Test Status
A one-click report that auto-drafts daily test status for sprint releases, showing pass, block, risk assessment metrics.
Test Completion Reports
End-of-sprint automated test completion reports tracking pass/fail/blocked ratios and key risk areas.
Accessibility Metrics
Live WCAG testing data and historical compliance trends across all platforms.
Defect Analytics
Aggregated defect analysis reporting for key metrics and defect aging patterns across teams.
AI-Assisted Report
Leverages automated LLM-powered caption/summary generation that adapts to different reporting contexts and requirements.
Automated Email Distribution
Blast-sends HTML or adds-on automated email distributions via SMTP with customizable recipients.
Persistent Form State
Form fields are Cosmos DB-backed — users can resume where they left off any time.
Enterprise Authentication
Full office SSO with Okta auth chain — enterprise-ready per-team permissions with role-based access control.
My Contributions
Designed Spring Boot 3.4 solution with Cosmos API, Java 21 design patterns — report forge architecture
Developed complete Spring Boot 3.4 modules with Java 21 + SMTP service delivery layer
Integrated AI/RAG Suite for real-time context-aware LLM-driven report generation
Implemented Kubernetes manifests for spec-driven multi-container deployments across environments
Built React 19 auto-save as Parity — report forge, persistence, full-stack writing
Migrated/refactored Cosmos DB with publish-key design for in-memory entities
Automated Playwright HTML and Headscrape module report generation for screenshot service
Implemented Top Veteran 2023 cert with Contrib.ai + prod and chained test suites
Pioneered e2e support: test-data infused-retry and fault tolerance
Co-created with Cluster observance and k8s-native CICD pipelines
Engineering Challenges
AI-Powered Structured Report Generation
Largest sprint-of-sprints task: feed raw test engineering metrics — 800 tests, block counts, fail analysis — into AI models to produce consistent, readable reports. Used function-calling with OpenAI models to extract tables then Thymeleaf to render formatted outputs.
Prompt Validation & Guardrails
Had a problem: open free-form can create multi-line input in services — uncertainty payload injection through user system-line shared terms and classes. Put in services around prompt design: ROLE-ID, RESPONSE-ID, and TERRAIN-ID, generated at the service layer. Responses are validated against JSON schema before being consumed.
Cross-Client HTML Email Rendering
Building email templates that works across clients (Outlook 2007, 2010, Gmail, Apple Mail, YMail) is painstaking but critical. Produced table-based HTML via email-specific Thymeleaf templates with conditionals: darling content display, accessibility metrics, and data-driven elements. Built dark/true-brand principles templates, HTML inline styling with CSS, and topic links for auto active screenshots.
Enterprise Security & Authentication
Integrated Mart to Tire Platform MTTP via MSGraphMiddleware with Okta-ID flows. The Platform proxy/interface is multifield, session-local-management, no role-based. Main functions: Scoped JWT with claims/RBAC patterns — session for user accounts — conditional bearer OIDC with API support against multi-tenant targets. Service configuration (Luca, credential) as externalized via containerized address.
Multi-Source Data Aggregation
Each external integration surface is aggregated data from a REST/Graph server at SERVICE ASSEMBLY (REST), accessibility scanners, and Jira APIs. The API-level per-page/API interface to aggregate the full concurrent data across apps on Monday skip any Monday use via APIs. We had to handle upstream service failures gracefully.
Automated Deployment Pipeline
Complete pod-full oriented deployment with Docker and Kubernetes to Blue-Green CICD pipelines (pipelines-to CICD pall), combo unit, integration, and safety-deployment. Spring boot Scanner is under-multi cross-references to Kubernetes containerized/service probes. The build for its command fixed-builds. Enterprise-specific configuration (snapshots) via internal image from local source to production.
Lessons Learned
AI outputs need strict contracts, not just prompts
LLMs return valid JSON/structures in-bulk. By producing JSON/JSON Schemas on all projected outputs and structured parsing with retry-on-fail, tables that key-conform always match the rendered template interface.
Design for the email client, not the browser
HTML email rendering is wildly inconsistent: prefer tables, Outlook, Inline/Cel, and Avoid Mso Styling. Thymeleaf templates with inline-styles, table-based layouts, and in-block CSS ensure reliable/preview output everywhere.
Partition key design in Cosmos DB matters more than you think
Querying HTTP 429 at the partition key-vs-indexed-id/head cross-partition queries for hot-partition data. Use items-as-data to improve horizontal batch-optimized query reads with RU/s throttling as priority for Cosmos.
Enterprise SSO integration requires early planning
The biggest Cosmos CORS-token-error over-custom in token-refresh-to-tng, assert-forwards, state-ful/token-ize scheme — flag optional for structured cleanup. Retrofitting SSO into a running app is significantly harder.
Preview endpoints save more time than you'd expect
Adding /api/v1/report/preview endpoints for real-time-view templates with first-user-data-entry fixtures along-with actively waiting — just real/full reset features via template changes.
Ship incrementally, validate with users
Releasing a brief report-sprint-as in first-feature → in-iteration → accumulating configuring/render after user feedback. Most XP is building the actual skills, file the test inputs for building on functional metrics.