Schedule Risk Agent
- Contest AWS Agentic MSP Hackathon (Devpost)
- Track Deliver AI-Driven Outcomes for Customers
- Status Submitted · judging 5–22 January 2027 · results around 28 January
See what changed, what matters, and what needs attention, in one evidence-backed schedule report. Schedule Risk Agent compares this week's construction schedule with last week's and tells the project team which tasks slipped, which became critical, and what to check first.
The video plays from YouTube when you press play. · Entry on Devpost ↗
Why we built it
Every week, construction teams export a new Microsoft Project or Primavera P6 schedule. Hundreds of linked activities move a little; one of them quietly loses its last day of float and the completion date slips. Large contractors pay planners and specialist tools to catch this. Smaller firms often have no planner, so nobody looks until the delay is visible on site.
The same firms already pay an IT service provider (an MSP) to run their IT. Schedule Risk Agent lets that provider add a weekly schedule review to the contract and run it for many customers at a known cost per report. Built for construction, reusable across project-driven industries, delivered by MSPs.
What it does
- For the customer: a private portal. They drop this week's schedule; the service compares it with last week's and returns a report: finish movement, activities that entered or left the critical path (with float before and after), failed schedule-quality checks, a 3-week look-ahead and actions for named owners.
- Evidence, not guesses: an Evidence table traces every number to the source file or a fixed calculation, cites activity IDs and writes N/A instead of guessing. The agent's own interpretations are labelled as such.
- An agent that investigates: after the overview, the agent decides which risks, if any, need a closer look, and checks what drives each one: a date constraint, the chain it drives, or another path meeting it at the same point. It never states a cause as fact, and two fixed checks catch any sentence that does and any number copied wrong.
- For the MSP: an operations console to onboard customers, run reports, see each run's time and cost, and a weekly job that reports on every customer with a new schedule.
How it's built
Deterministic analysis comes first: the schedule file is read with the open-source MPXJ library, and dates, float and critical flags come from the scheduling tool's own values. Then an agent built with Strands Agents runs on Amazon Bedrock AgentCore Runtime, picks what to check with its schedule tools, and Claude on Amazon Bedrock writes the report. AgentCore Memory keeps each customer's history, and AgentCore Gateway offers the same tools over MCP for an MSP's own copilot. Everything is deployed by CloudFormation from GitHub Actions.
What we measured
- Accuracy of the numbers: in the 15 of 20 public Primavera P6 files that store P6's own results, all 1,604 finish dates and 792 float values we read match the stored values. On 44 public Microsoft Project files, 1,874 comparable tasks showed no mismatch in dates, baselines, progress or slack.
- Does the investigation help? In a blind test (5 scenarios × 5 runs, graded by a separate Claude model against an answer key written first), causes stated without evidence fell from 15 to 1 compared with the first version.
- Speed and cost: 24–36 seconds per report on schedules of 100 to 1,000 activities, at an estimated $0.05–0.15 in AWS run cost per report.
These are our own tests on public and synthetic schedules, not customer validation. It reports and explains risk; it does not predict the probability of a delay.
한국어 소개 In Korean
Schedule Risk Agent는 LRLR이 AWS Agentic MSP Hackathon(Devpost)에 낸 출품작입니다. 제출을 마쳤고, 심사는 2027년 1월 5일부터 22일까지, 결과 발표는 1월 28일 전후입니다.
건설 현장은 매주 Microsoft Project나 Primavera P6 공정표를 새로 뽑습니다. 수백 개 작업이 조금씩 밀리다가, 그중 하나가 여유일을 다 써 버리면 준공일이 밀립니다. 큰 건설사는 공정 담당자와 전문 도구로 이걸 잡지만, 작은 회사는 공정 담당자가 없어서 현장에서 지연이 보일 때까지 아무도 모르는 경우가 많습니다.
Schedule Risk Agent는 이번 주 공정표를 지난주와 비교해 준공일이 얼마나 움직였는지, 어떤 작업이 새로 주공정에 들어왔는지, 어떤 점검 항목을 통과하지 못했는지 정리한 보고서를 씁니다. 모든 숫자는 원본 파일이나 정해진 계산으로 추적되고, 알 수 없는 건 추측하지 않고 N/A로 씁니다. 에이전트는 더 볼 필요가 있는 위험만 골라 무엇이 그 작업을 붙잡고 있는지 확인하지만, 원인을 사실처럼 단정하지는 않습니다.
이 서비스는 이미 고객사의 IT를 맡고 있는 관리 서비스 업체(MSP)가 매주 여러 고객사에 돌리는 형태로 설계했습니다. 보고서 한 건에 24–36초, AWS 비용은 약 $0.05–0.15입니다. 공개 P6 파일 20개 중 P6 계산값이 저장된 15개에서, 읽어 낸 종료일 1,604개와 여유일 792개가 저장된 값과 모두 일치했습니다.
자세한 내용은 3분 데모 영상과 Devpost 출품 페이지에서 볼 수 있습니다.