Agentic AI on AWS

Synapxe Healthcare — Agentic AI Workshop Programme

Two half-day sessions · Foundations for all staff, then an optional Use Case Deep Dive · No coding required · Built for hospital administration, finance & agency teams

A Two-Part Programme

The workshop is delivered as two half-day sessions. Foundations takes all staff from understanding what an agent is to building one. The optional Use Case Deep Dive is for those who want to mature their own idea into a working, verified assistant with a plan to implement.

📖

Pre-read · Self-paced

How AI Works — short interactive explainers (tokenization, embeddings, inference) sent before Session 1. Not live workshop time.

🌱

Session 1 · Foundations (½ day)

All staff. From chatbots to agents → prompting → build the Guideline Assistant on AgentSea → design your own agent on the canvas.

⚙️

Session 2 · Deep Dive (½ day)

Optional follow-up. Bring your canvas or a new problem → workflow patterns → robust build + verify → automation plan → implementation path.

🔗 The through-line: Foundations ends with an Agent Design Canvas for a workflow from your own team. That canvas is the ticket into the Deep Dive, where it's refined, built properly, and prepared for hand-off to the DNA / IT team.

Who Is This For?

🏥

Hospital Administration

Administrative staff managing documents, correspondence, and operational workflows across agencies

💰

Finance Teams

Budget analysts, procurement officers, and finance staff handling reports, vendor assessments, and compliance

📋

Agency & Operations

Ministry and agency staff managing policies, meeting coordination, and cross-department communications

👥

HR & People Teams

HR professionals handling leave processing, onboarding documentation, and staff communications

What is AgentSea?

AgentSea is Synapxe's internal AI platform — your starting point for AI-powered productivity. Built on Amazon Bedrock, it lets you deploy AI agents for your daily work. As you grow, you'll also explore Bedrock AgentCore, Strands Agents, and Kiro IDE — the AWS agentic AI stack that powers and extends what AgentSea offers.

📄

Document Analysis

Upload PDFs, Excel, or Word files — the agent reads, summarizes, and extracts key information

✍️

Document Generation / Builder

Create reports, letters, summaries, and brand-compliant documents from your instructions

📧

Outlook Integration

Read and summarize emails, check calendars, and draft responses

🚀 Your AI journey: Start with AgentSea for daily productivity → understand Amazon Bedrock as the foundation powering it all → for technical teams, build with Bedrock AgentCore, Strands Agents, and Kiro IDE. Connectors (knowledge spaces, SharePoint), multi-step workflows, and shareable skills are shown as concepts, subject to platform readiness.

Prerequisites

  • No coding experience required — this is a conceptual, design, and hands-on build workshop
  • Complete the self-paced How AI Works pre-read before Session 1
  • Curiosity about how AI can improve your team's workflows
  • Familiarity with your own team's daily processes (you'll design an agent for one of them)
  • Laptop with internet access and a modern browser · AgentSea platform access (provided)
  • Deep Dive only: completed Foundations · bring your Agent Design Canvas, or a new problem statement
⚠️ Data Privacy Notice: Do not include any confidential patient information, personally identifiable information (PII), or sensitive clinical data in any exercises. All sample data in this workshop is synthetic. Follow PDPA guidelines at all times.

🌱 Foundations — Half-Day Workshop

A hands-on, half-day session that takes all staff from understanding what an AI agent is to building one. Everyone builds a shared, grounded assistant together — the Guideline Assistant — then designs an agent for their own work on the Agent Design Canvas, the springboard into the optional Use Case Deep Dive.

⏱️

~4 hours

Half day — runs as a morning or afternoon session

👥

All staff

Non-technical · healthcare administration context

📖

Pre-read

Self-paced How AI Works before the session

📄 Formal outline: View the proposal outline → (pending customer / leadership approval)

Learning Outcomes

By the end, participants can:

  • Explain what makes an AI system an agent (versus a chatbot) — and why that shift matters for their work
  • Write a clear system prompt using the 4 pillars + persona to steer an agent
  • Build a simple working agent on AgentSea using Document Analysis and the enhanced Document Builder (brand-compliant output)
  • Design an agent for their own workflow using the Agent Design Canvas — captured as a shareable one-pager
  • Recognise the common agentic workflow patterns and where AgentSea is heading

Agenda

TimeModuleDescription
09:00–09:10F0 · Welcome
from AI users to agent builders
Goals, quick poll, and meet the shared use case (the Guideline Assistant)
09:10–10:00F1 · From Chatbots to AgentsWhy a chatbot hits a wall, the four ideas that make an agent, and the levels of autonomy
10:00–10:40F2 · Prompting your agentThe 4 pillars + persona → writing a clear system prompt
10:40–10:55Break
10:55–11:55F3 · Build the Guideline AssistantGuided build on AgentSea: ask an internal guideline → grounded answer → brand-compliant briefing (Document Builder)
11:55–12:45F4 · Agent Design Canvas
your use case
Design an agent for your own workflow → a branded HTML one-pager; the bridge to the optional Deep Dive
12:45–13:00F5 · Wrap-up & what's nextKey takeaways, the optional Use Case Deep Dive, and the champion network

Timings are adjustable; runs equally well as a morning or afternoon session.

The Two Signature Activities

🌊

F3 · Build the Guideline Assistant

A light, guided build everyone completes together. Give the agent a skill (Document Analysis) and a system prompt (answer only from the document, cite the section, say "not covered" if it isn't there), ground it in a synthetic guideline, test a covered vs a deliberately uncovered question, then produce a brand-compliant briefing with Document Builder. Teaches the single most important idea: an agent grounded in your documents, not the open internet.

Open the F3 lab →
🤖

F4 · Agent Design Canvas

Using the Guideline Assistant as the template, each participant designs an agent for a task in their own function (Finance, HR, Operations, Procurement, admin). The canvas captures purpose, steps, the skills needed, guardrails, and how they'd know it's working — captured as a Markdown file, then a branded HTML one-pager to take away. This canvas is the ticket into the Use Case Deep Dive.

Open the Canvas →

Starter Prompts

Copy-paste starting points for F2–F4. Sharpen a vague ask into a powerful one, then shape it into a deployable system prompt.

Sharpen a prompt (F2)

❌ BEFORE — Vague prompt
Summarize this document for me.
✅ AFTER — Powerful prompt
You are a Senior Procurement Analyst with 10 years of experience in healthcare vendor assessment. Read the attached vendor proposal and produce: 1. A summary table with columns: Vendor Name, Service Offered, Contract Duration, Total Cost (SGD), Key Risks 2. A list of 3 strengths and 3 concerns about this vendor 3. A recommendation: PROCEED / REVIEW FURTHER / REJECT — with one-sentence justification Rules: - Only use information from the attached document — do not make assumptions - Flag any missing information as "NOT PROVIDED" - All costs in SGD - Keep the summary under 500 words

System prompt template (F3 / F4)

SYSTEM PROMPT TEMPLATE
You are [ROLE — use persona formula]. Your purpose is to [ONE SENTENCE — what this agent does]. ## What You Do When the user uploads [INPUT TYPE], you will: 1. [STEP 1 — what to extract/analyze] 2. [STEP 2 — what to produce/generate] 3. [STEP 3 — what format to output] ## Output Format [Describe the exact format: table, bullet points, report sections] ## Rules You MUST Follow - Only use information from the uploaded document(s) - If information is missing, state "NOT PROVIDED" — never guess - All monetary values in SGD - Keep responses under [WORD LIMIT] words - Use professional, clear language ## You Must NEVER - [NEGATIVE CONSTRAINT 1 — e.g., "provide medical advice"] - [NEGATIVE CONSTRAINT 2 — e.g., "share patient identifiable information"] - [NEGATIVE CONSTRAINT 3 — e.g., "approve amounts above $50,000 without flagging"] ## When to Escalate Say "⚠️ Please consult your supervisor" when: - [ESCALATION TRIGGER 1] - [ESCALATION TRIGGER 2] - You are unsure about the correct interpretation

⚙️ Use Case Deep Dive — Half-Day ILT

A hands-on follow-up for staff who identified a promising use case and want to take it further. Where Foundations builds awareness and a first idea, this Deep Dive is about refining and implementing it — deeper agentic patterns, a robust agent design, a full build with verification, and a clear path to hand over to the DNA / IT team.

⏱️

~4 hours

Half day — morning or afternoon

🎟️

Optional

For a subset who want to mature a use case

🤖

Bring your canvas

Continue from Foundations, or bring a new problem

📄 Formal outline: View the proposal outline → · Prerequisite: completed the Foundations half-day

Learning Outcomes

By the end, participants can:

  • Select the right agentic workflow pattern for their use case from the four core patterns
  • Write a robust skill / system prompt with guardrails and escalation rules for their use case
  • Build and iterate a working assistant on AgentSea, including a verification step that catches errors
  • Design an agentic automation plan — triggers, steps, and human-in-the-loop checkpoints — ready for when AgentSea's Workflow Builder lands
  • Complete an Agent Design Canvas and prioritise their use case on value, feasibility and risk
  • Define measurable success metrics and an implementation path — what to hand to the DNA / IT team

Agenda

TimeModuleDescription
09:00–09:15D0 · Welcome
bring your use case
Frame your use case — continue from your Foundations canvas or bring a new problem
09:15–09:55D1 · Agentic workflow patternsAll four patterns (Chaining · Parallelisation · Routing · Orchestration) — matched to your use case
09:55–10:35D2 · Build skills & instructionsHands-on: craft the system prompt and configure the skills your agent needs — guardrails, escalation, what NOT to do
10:35–10:50Break
10:50–11:45D3 · Build & verify (capstone)Build a working assistant on AgentSea; the deliberate-error verification / trust-but-verify moment; iterate
11:45–12:15D4 · Agentic Automation (design)Concept + design: triggers, approval nodes, human-in-the-loop → produce an automation plan (Markdown)
12:15–12:50D5 · Use-case refinement
Agent Design Canvas
Full canvas: value / feasibility / risk, success metrics, and the path to implementation
12:50–13:00D6 · Showcase & next stepsShare designs, implementation roadmap, and how to engage the DNA / IT team

Timings are adjustable; runs equally well as a morning or afternoon session.

Worked Examples by Function

Anyone who prefers a ready-made starting point can adopt one of these instead of their own idea.

📑

Vendor Proposal Analysis

Procurement / Finance — extract terms, compare vendors, flag risks, recommend

📜

Policy & Circular Compliance

Governance — check a document against a policy or circular, cite the clause

🗂️

Claims / Leave Processing

HR — apply the rules in a staff guideline to a request and draft the response

📈

Budget Variance Commentary

Finance — read the numbers, explain the variance, draft a memo

D3 · The Build & Verify Capstone

On current AgentSea (single agent + Document Analysis + Document Generation + system prompt):

  • Deploy the system prompt drafted in D2 for your use case
  • Upload a document pack; run Document Analysis → structured output
  • Iterate on a deliberate-error document — the trust-but-verify teaching moment; refine the prompt until the agent catches it
  • Generate the final output with Document Builder — a brand / template-compliant deliverable (a vendor comparison in Excel, a formatted variance memo/PPT, or a policy briefing) — and review quality
🛡️ Trust but verify: AgentSea handles Restricted and Sensitive (Normal) information — and you always verify responses. The verification step is the heart of this capstone: the agent does the heavy lifting, a human stays accountable for the result.

D4 · Agentic Automation — Design Only

The shift from an assistant you invoke to an agent that runs on a trigger or a schedule — the L1 → L2 → L3 autonomy ladder. Three questions frame any automation: What triggers it? · What runs? · Where does a human stay in the loop?

This module is design-only — no hands-on build. The deliverable is an automation plan (Markdown): the trigger, the steps, the human checkpoints, and the fallback — ready to hand to the DNA / IT team.

Resources & Reference

📖 Quick Reference Cards

🔗 AgentSea Platform

🎓 Continue Learning

📨 The Learning Pathway

From Foundations to implementation:
  • Before Session 1: complete the self-paced How AI Works pre-read
  • Session 1 — Foundations: build the Guideline Assistant, then design your own agent on the canvas
  • Between sessions: try your canvas on AgentSea — deploy your system prompt, refine guardrails and output
  • Session 2 — Deep Dive (optional): mature your use case into a verified build + automation plan + implementation path
  • Ongoing: join the AI Champion network; hand mature use cases to the DNA / IT team