Highlander — SRD-10 Dashboard

SRD-10 · Incident Investigation & Corrective Action System · Mar 27–29, 2026

Highlander

Built solo during a 59-hour hackathon by coordinating a multi-agent AI team — TPM, two SWEs, and QA — across a Flutter + Go stack. One developer. Four agents. 281 commits. Zero open issues.

Claude Opus + Sonnet Flutter · Go ADA / WCAG Playwright · 542 tests RAG · AutoResearch MCP · JSON-RPC 2.0

Quick links

Live App GitHub Repo Ask the Team (GPT)

By the numbers

Build time

59h

Mar 27 08:35 → Mar 29 19:18

Commits

281

main branch

Branches

all merged

Pull requests

93 merged · 0 open

Issues

103

all closed

Build phases

phases 0–11

Test cases

542

15 Playwright suites

AI agents

TPM · SWE-1 · SWE-2 · QA

Lines of code

69,282

across all components

API endpoints

76+

registered handlers

Docker services

backend · web · pg · ollama

Flutter modules

feature modules

Lines of code breakdown

Component	Files	Lines
Flutter / Dart frontend	113	39,452
Go backend	65	13,268
Playwright tests	15	10,544
Documentation markdown	25	6,018

Contributor breakdown

claude (AI agents) #1 commits

249

commits on feature branches

73,500

7,700

lxRbckl (human) #1 lines

183

commits on feature branches

109,285

6,406

Total lines changed

Commit split (432 total)

claude (AI)

249 (58%)

lxRbckl

183 (42%)

Claude Max usage — 59-hour build window

Token consumption by day

Mar 27

430M

tokens

Mar 28

953M

tokens — peak day

Mar 29

410M

tokens

AI optimization — AutoResearch

AutoResearch Automated RAG wiki pipeline

AutoResearch was used to systematically improve the in-app AI assistant by auto-generating a structured wiki from the codebase on every commit. Rather than manually tuning prompts, the eval pipeline measures accuracy before and after each update — creating a continuous optimization loop. The same pattern Shopify used for a 53% improvement across 120 experiments.

10%

baseline accuracy

88%

with RAG wiki

+79pt

net lift

Eval results by domain (42 test cases)

Baseline

10%

With wiki RAG

88%

100%

Routes & RBAC

100%

Accounts

100%

Incidents

100%

OSHA compliance

100%

Tech architecture

+79pt

net accuracy lift

Issue breakdown

Enhancement

74 (72%)

Bug

26 (25%)

Documentation

3 (3%)

Difficulty distribution

Trivial

Routine

Complex

Critical

Agent team

TPMTechnical Program Manager

Orchestration, GitHub Issues, architecture decisions, model routing — Sonnet for routine, Opus for complex.

task decompmodel routingtrade-offs

view prompt · tpm.md

SWE-1Full-Stack Developer

Flutter UI, Go API, database models, offline sync. Owned 52 feature branches.

fluttergo api52 branches

view prompt · swe-1.md

SWE-2Full-Stack Developer

Parallel feature track — distinct ownership across the feature set. 39 feature branches merged.

fluttergo api39 branches

view prompt · swe-2.md

QAQuality Assurance

542 Playwright test cases across 15 suites. Tested in isolated git worktrees — no persistent branches needed.

playwrightada/wcaggit worktree

view prompt · qa.md

Branch ownership

SWE-1

52 branches

SWE-2

39 branches

0 (worktree)

Build phases — 59 hours

Foundation

Set up auth, navigation shell, and admin settings as the base for all features.

3 tasksbuild-plan.md

Incident Reporting

Build the incident data model, API, and full reporting UI with OSHA determination and railroad notifications.

2 tasksbuild-plan.md

Investigation + CAPA

Add investigation lifecycle — 5-Why, contributing factors, witness statements — and CAPA management with verification workflows.

4 tasksbuild-plan.md

Dashboard, Recurrence, Audit, Notifications

Build the safety dashboard with TRIR/DART metrics, incident recurrence linking, audit log viewer, and escalation notifications.

4 tasksbuild-plan.md

Hardening + Polish

Lock down RBAC across all endpoints, seed realistic demo data, and run full integration testing.

3 tasksbuild-plan.md

Differentiators

Add the in-app AI chat assistant, keyboard shortcuts, and AI-driven page navigation with form filling.

3 tasksbuild-plan.md

Future Roadmap

Implement deferred rubric items: real login system, offline mode, fishbone diagrams, auto-recurrence detection, advanced analytics, OSHA exports, email notifications, and training verification.

7 tasks — largest phasebuild-plan.md

Judge Differentiators

Add PDF exports, global search, incident timeline, dark mode, role-based landing pages, and live activity feed.

6 tasksbuild-plan.md

Domain Innovation

Build the incident map view with GPS markers, voice-to-text reporting, and dashboard PDF summary reports.

3 tasksbuild-plan.md

Demo Polish

Add the onboarding coach-mark tour and real-time WebSocket updates for instant notifications.

2 tasksbuild-plan.md

AI-Powered Navigation

Enable query parameter pre-fill on forms and AI chat URL generation so the assistant can create clickable deep links.

2 tasksbuild-plan.md

MCP Agent Integration

Build the agent API key system, capabilities endpoint, full MCP server protocol, and agent session monitoring.

4 tasksbuild-plan.md

Tech stack

Claude Opus + Sonnet AutoResearch Flutter / Dart Go + GORM PostgreSQL Ollama / Qwen 2.5 3B MCP JSON-RPC 2.0 bcrypt + JWT Playwright GitHub Actions Docker Compose WCAG 2.1 AA Azure AD-ready

Generated 2026-03-29 · lxRbckl/highlander