BDI Hackathon 2026 · AWS Sponsored · Room 3: ก่อนเกิดภัย

SinkAlert × AWS

Migration Strategy & Architecture Plan
Leveraging Amazon Bedrock, IoT Core, SageMaker & Kiro for Maximum Hackathon Impact

📋
Executive Summary
Why AWS Migration Matters for Our Score
BDI Hackathon is sponsored by AWS. Using AWS services is not optional — it's a scoring multiplier. The AWS AI Portfolio document (80 pages) reveals exactly what judges want to see: Amazon Bedrock AgentCore for agent architecture, SageMaker for ML pipelines, and Kiro for spec-driven development. Teams that ignore the sponsor's stack leave points on the table. Teams that embrace it demonstrate ecosystem alignment (Dimension 7, 10pts) AND technical maturity (Dimension 2, 20pts).

Score Impact — Per Judging Dimension

#DimensionWeightCurrentWith AWSWhat AWS Adds
1Problem Fit & Use Case201418 AWS IoT architecture shows real-world deployment; KVS for dashcam integration
2PoC Progress & Demo201217 Bedrock AgentCore demo = production-ready; SageMaker notebook = training proof
3Data & AI Approach151114 Bedrock multi-model + SageMaker XGBoost + Rekognition CV = AWS-native AI stack
4Validation & Evidence1057 Bedrock Guardrails for safety; CloudWatch for monitoring
5Impact, ROI & Social Value151214 Serverless cost model (Lambda); AWS Free Tier for pilot; transparent pricing
6Team & Learning1068 Kiro adoption shows AWS tool proficiency; spec-driven development discipline
7Ecosystem & Strategic Fit1069 100% AWS-native: Bedrock + SageMaker + IoT + Lambda + KVS = runs on sponsor's platform
TOTAL1006687+21 points from AWS alignment
🏗️
Target Architecture — AWS-Native SinkAlert
Each component maps to a specific AWS service from the BDI portfolio
🛰️ Data Ingestion Layer
AWS IoT Core + Kinesis Video Streams
Camera feeds → KVS secure streaming. IoT sensors → MQTT telemetry. Edge CV on device. Sentinel-1 InSAR → S3 batch ingest. All data lands in S3 data lake.
🧠 AI/ML Layer
Amazon Bedrock + SageMaker
Claude 3.5 Sonnet for Thai-language reasoning. Nova for multimodal (satellite images). SageMaker for XGBoost training. Rekognition for CV augmentation. Bedrock Knowledge Bases for RAG over Thai disaster docs.
🤖 Agent Layer
Bedrock AgentCore
AgentCore Runtime for multi-agent orchestration. AgentCore Memory for persistent context. AgentCore Identity for secure tool access. AgentCore Observability with CloudWatch traces.
🚨 Alert & Action Layer
AWS Lambda + IoT Events
Lambda functions trigger alerts based on XGBoost risk scores. IoT Events detect anomaly patterns. SNS/SES for multi-channel notifications (LINE, SMS, Email).
🗺️ Visualization Layer
AWS Amplify + CloudFront
MapLibre GL JS dashboard hosted on Amplify. CloudFront CDN for global low-latency. Cognito for user authentication. Real-time WebSocket updates via AppSync.
🛡️ Safety & Governance
Bedrock Guardrails
88% harmful content filtered. PII redaction for public alerts. Automated Reasoning checks: 99% hallucination accuracy. Groundedness verification on all model outputs.

Complete Data Pipeline (AWS-Native)

📹 Dashcam / CCTV
Kinesis Video Streams
🔍 Edge CV
IoT Greengrass
📦 Raw Storage
S3 Data Lake
🛰️ InSAR Processing
SageMaker Processing
🤖 Risk Scoring
SageMaker XGBoost
🧠 Thai Analysis
Bedrock Claude 3.5
🚨 Alert Dispatch
Lambda + SNS
🗺️ Dashboard
Amplify + CloudFront
🔑
Key AWS Services — What We Use & Why
Extracted from BDI×AWS AI Portfolio (pages referenced)

1. Amazon Bedrock (p.15–18) — Foundation Models

ModelUse CaseWhy This ModelCost
Claude 3.5 SonnetThai-language risk reasoning, alert generationBest Thai reasoning among Bedrock models. Extended access free for hackathon evaluation.$6/M input · $30/M output tokens
Amazon Nova ProMultimodal: satellite image + text analysisAWS-native. Processes InSAR visuals + Thai descriptions together.AWS on-demand pricing
NVIDIA Nemotron Super 49BResearch synthesis, 1M context window for reading papersAlready using via NVIDIA NIM. Available on Bedrock Marketplace (p.18).Marketplace pricing
DeepSeek R1Complex root cause analysisAlready using via DeepSeek API. Also on Bedrock Marketplace.Marketplace pricing

2. Amazon Bedrock AgentCore (p.20–31) — Production Agent Platform

Why this wins points: AgentCore is AWS's flagship agent platform presented at BDI Workshop #4. Using it shows you listened to the sponsor's technical direction.

ComponentSinkAlert UseScore Impact
AgentCore Runtime (p.22-23)Multi-agent: Supervisor agent (Strands) orchestrates InSAR agent + CV agent + Alert agentD2 (+3pts): Production-grade multi-agent architecture
AgentCore Memory (p.22)Persist risk assessments, incident history, Thai-language context across sessionsD3 (+1pt): Shows memory-aware AI design
AgentCore Identity (p.24-26)Secure access to AWS resources (S3, SageMaker) and external tools via OAuthD4 (+1pt): Secure, auditable agent operations
AgentCore Observability (p.27-28)CloudWatch traces, OTEL logs. Full audit trail of every AI decision.D2 (+2pts): Operational readiness visible to judges
AgentCore Gateway (p.21)Browser tool for automated data collection. Code interpreter for analysis.D3 (+1pt): Agent tooling maturity

3. Bedrock Knowledge Bases + Guardrails (p.19) — Safety for Public Alerts

FeatureSinkAlert UseProven Metric (from AWS)
Knowledge BasesRAG over Thai disaster documents: DMR sinkhole maps, DOH road specs, DDPM protocolsContextual retrieval with citation
Content FilteringBlock harmful/misleading alerts before they reach the public88% harmful content filtered
Automated ReasoningVerify model outputs against ground truth — prevent hallucinated sinkhole alerts99% hallucination detection accuracy
PII RedactionStrip personal data from incident reports before public dashboardsHIPAA-eligible PII masking
Groundedness ChecksEnsure every alert cites specific data sources (InSAR image, rainfall reading, road segment)Context-based verification

4. Amazon SageMaker (p.5) — Custom ML Training

CapabilityCurrent SetupAWS Migration
XGBoost TrainingLocal Python script (not trained yet)SageMaker XGBoost built-in algorithm → managed training job with DMR labels
InSAR ProcessingLocal MintPy + ISCE2SageMaker Processing job → containerized ISCE2 pipeline → outputs to S3
Model RegistryNoneSageMaker Model Registry → versioned models with lineage tracking
InferenceNot deployedSageMaker real-time endpoint → REST API for risk scoring
Hyperpod (p.5)N/AFuture: distributed training on Trainium for large geospatial models

🛣️ Training Data Strategy — Why Japan RDD2022

RDD2022 provides multi-national road damage images. We selected Japan COCO as our transfer-learning source because it's the only subset with Thailand-matching road degradation physics:

ConditionJapanThailandMatch
ClimateTropical monsoonTropical monsoon
Rainfall1,500–2,500mm1,200–2,400mm
Road surfaceAsphaltAsphalt AC 60/70
Damage typesAlligator, pothole, crackAlligator, pothole, crack

Norway (subarctic ❌), Turkey (dry ❌), USA (concrete roads ❌) were eliminated. Training on Japan's tropical monsoon crack patterns ensures transfer learning preserves relevant features before fine-tuning on Thai dashcam data.

5. AWS IoT + Kinesis Video Streams (p.6–10) — Edge to Cloud

Why this differentiates: The AWS AI Portfolio dedicates 5 full pages to IoT + KVS. This is a core AWS competency. SinkAlert positioning as an "IoT-connected road monitoring system" — even with simulated sensors — shows architectural vision.

ComponentSinkAlert UseDemo Potential
Kinesis Video StreamsDashcam/security camera ingestion → ML inference pipeline (p.10 diagram)Show a video feed → KVS → Rekognition detecting road anomalies live
AWS IoT CoreSimulated soil moisture sensors → MQTT telemetry → IoT Events triggerDashboard showing "live" sensor data feeding the risk model
IoT GreengrassEdge CV on roadside cameras — YOLOv8 runs locally, only alerts sent to cloudArchitecture diagram showing edge-to-cloud pipeline
IoT Device ShadowRoad segment "digital twin" with current risk stateMap click → device shadow shows full risk profile

6. AWS Lambda + Serverless — Cost Model for Judges

ServiceMonthly Cost (Pilot)What It Does
Lambda$0 (1M free requests/mo)Alert processing, data transformation, API endpoints
S3~$3/mo (50GB InSAR data)Data lake: satellite imagery, sensor telemetry, incident reports
Bedrock (on-demand)~$15/mo (est. 500K tokens)Claude 3.5 Sonnet for risk reasoning + Nova for multimodal
SageMaker (training)~$5/job (ml.m5.xlarge × 2hr)XGBoost training on DMR labels
Kinesis Video Streams~$2/mo (1 stream, 1 cam)Video ingestion for PoC demo
TOTAL Monthly (Pilot Phase)~$25/month
💡 Pitch Point: "SinkAlert costs ฿7/km to monitor Thailand's 68,000km of roads. On AWS, the entire pilot runs for $25/month. Preventing one Samsen-scale sinkhole (฿1.04B damage) pays for 1,386 years of SinkAlert operation."

7. Kiro — AI-Powered Development (p.52–78)

Kiro was featured at Workshop #3 and teams were explicitly instructed to download Kiro before the session. This is not optional — it's an expected tool.

Kiro FeatureWhat It DoesSinkAlert Use
Spec-Driven Development (p.52-53)Turn prompts into requirements, system design, discrete tasksGenerate SinkAlert architecture spec → Kiro agents implement scaffolding
Agent Hooks (p.54)Auto-trigger agents on file save — generate docs, tests, optimize codeAuto-generate documentation when model code changes
Steering Files (p.55-60)Markdown rules in .kiro/steering/ — version-controlled team knowledgeCode SinkAlert coding standards, AWS conventions, BDI hackathon rules
MCP Servers (p.61-65)Connect AI to Jira, GitHub, AWS, databasesConnect Kiro agents directly to AWS Bedrock + SageMaker
Agent Skills (p.70-73)Packaged task workflows (SKILL.md + scripts + references)Create "SinkAlert Model Training" skill — reproducible ML pipeline
Custom Agents (p.74-77)Specialized AI personas with scoped toolsDevelopment agent (full access) + Production agent (read-only alerts)
📅
4-Phase Migration Plan — Before Soft Pitch (Jul 10)
8 days. Focused on pitch-demo deliverables, not full production migration.
PHASE 1 · Jul 2–3
🔑 AWS Account + Bedrock Setup
  • Activate AWS account with hackathon credits
  • Request Bedrock model access (Claude 3.5 Sonnet, Nova)
  • Configure IAM roles for SageMaker, S3, Lambda
  • Set up Bedrock Guardrails for Thai-language safety
  • Create Bedrock Knowledge Base with DMR + DOH docs
PHASE 2 · Jul 4–5
🧠 SageMaker ML Pipeline
  • Upload DMR sinkhole labels + InSAR features to S3
  • Create SageMaker notebook: XGBoost training script
  • Run first training job → validate F1 score
  • Register trained model in SageMaker Model Registry
  • Deploy inference endpoint for live risk scoring
PHASE 3 · Jul 6–7
🤖 AgentCore + IoT Demo
  • Set up Bedrock AgentCore agent (Strands SDK)
  • Create multi-agent: Supervisor + InSAR + CV + Alert
  • Simulate IoT sensor data → Lambda → Bedrock reasoning
  • Demo: "Show me Bangkok road risk" → agent calls SageMaker + Bedrock
  • Configure AgentCore Observability for CloudWatch dashboards
PHASE 4 · Jul 8–9
🎤 Pitch Polish + Kiro
  • Build final pitch deck in PDF (submit by Jul 9, 22:00)
  • Record video demo: AWS console walkthrough
  • Set up Kiro with SinkAlert project + steering files
  • Create Agent Skill for model retraining workflow
  • Practice 5-min pitch with AWS architecture slides

⚠️ Risk Register

RiskMitigation
AWS credits not activated in timeUse AWS Free Tier ($0) for all PoC services. Bedrock on-demand models incur cost only when called. Lambda free tier: 1M requests/month.
Bedrock model access delayRequest Claude 3.5 Sonnet Extended Access immediately (12-24hr approval). Fallback: use Amazon Nova (instant access, no approval needed).
SageMaker training results poorPitch the pipeline, not the accuracy. "XGBoost V1 trained on DMR labels — score improving weekly as we add InSAR features." Honesty > fake numbers.
AgentCore too complex for 8 daysDemo the simplest agent flow: User asks → AgentCore calls SageMaker endpoint → Bedrock explains result in Thai. 30-min setup.
🎤
Pitch Integration — Where AWS Appears in Your 5 Minutes
Every judge must hear "AWS" in at least 3 of 6 topics
#Topic (5 min total)AWS MentionTime
1ปัญหา & กลุ่มเป้าหมาย "AWS IoT + KVS ingest dashcam data from 68,000km — the same architecture on p.10 of the AWS AI Portfolio" 40s
2ทางแก้ & Value Proposition "Running on Amazon Bedrock AgentCore — cost ฿7/km, $25/mo pilot. 1,386 years of operation = cost of preventing one Samsen" 50s
3Data & AI Approach "SageMaker XGBoost on DMR labels. Bedrock Claude 3.5 for Thai reasoning. Bedrock Guardrails: 99% hallucination detection. Rekognition CV augmentation." 60s
4PoC Progress "Training running on SageMaker. AgentCore demo live at dashboard.cfoth.ai/sinkalert/. Built with Kiro AI-powered IDE — spec-driven development." 50s
5Feasibility & Validation "AWS Free Tier covers entire pilot. S3 stores all Thai government data. Lambda handles 1M alert requests/month at $0." 40s
6ทีม & Next steps "Post-Soft Pitch: migrate to Bedrock AgentCore production. Enroll in AWS Mentoring Hour. List on AWS Marketplace for Thai government procurement." 40s