Forward Deployed Engineering · Avanade Inner Source

A live AI agent demo,
in under 20 minutes.

Built for FDEs who are tired of arriving at pre-sales calls with slides instead of something the customer can actually touch. Personalised to their systems, their language, their questions — in real time.

Get the repo → See how it works

FDE — Forward Deployed Engineer · Avanade

fde-demo-worlds — localhost:5173

$ npm install && npm run dev

✓ Ready on http://localhost:5173

STEP 1 Open /configure → Paste from LLM → copy generation prompt

STEP 2 Describe the customer's use case in Claude or Copilot, paste response back

STEP 3 Agent loads in gallery — click Start demo. Customer is interacting.

✦ Demo live · elapsed: 00:03:47

The gap we're closing

The demo you could show
vs the deck you're showing

✗ Generic demos without the client's system names don't land. "Imagine Workday here" is not a demo.
✗ Building a convincing personalised demo from scratch takes days, not hours — time you don't have.
✗ FDEs spend time on scaffolding and APIs instead of the narrative and discovery that actually close deals.
✗ After the demo, the SA still needs a separate briefing to scope — discovery data lived in your head, not a format they can use.

I built FDE Demo Worlds after sitting in too many pre-sales calls where the conversation was happening in slides rather than in something the customer could touch. The tool grew out of a recurring pattern — a gap between what sellers need to show and what our solution architects need to scope.

The goal: an FDE with minimal information about a client's problem can spin up a live, interactive demo in 3–10 minutes. The customer types their own questions. They get real streamed responses. After the session, the tool produces two artefacts that close the loop straight into the ROM/AEP process.

How it works

From cold start
to customer conversation

No code. No backend. No shared infrastructure. Runs entirely on your laptop, against your own API key.

01 · SETUP

One-time install, under 5 minutes

Clone the repo, npm install, drop your API key in .env. Set your name in platform.config.json. Done. The reference HR agent loads immediately — you can demo it right now.

02 · BUILD

Generate a personalised agent in minutes

Open /configure, choose Paste from LLM. Copy the generation prompt, take it to Claude or Copilot, describe the client's systems and use case. Paste the response back. Agent appears in the gallery. That's it — no JSON editing required unless you want to tune it.

03 · DEMO

The customer sees a pixel-accurate Teams UI

They interact with an agent that knows their systems, speaks their language, and handles their specific use case. Responses stream in real time. Suggested follow-up prompts guide the conversation. Reference cards show which document a response came from. Action cards confirm completed tasks.

04 · CAPTURE

Discovery data goes straight to scoping

After the demo, the FDE Feedback modal walks through 21 discovery framework areas — purpose, users, integrations, build complexity, ROI, governance, and more. Export two artefacts: a client questionnaire to share with the customer, and an FDE Design Report that feeds directly into the SA's ROM/AEP session.

Agent gallery

All your demos, in one place

Each agent lives in a folder. Adding a folder adds a demo — the shell never changes. The gallery shows every agent you've built, with language launch buttons so you can flip between EN and the customer's local language mid-session.

The FDE's name and company personalise the gallery header automatically from platform.config.json.

folder-based hot reload multilingual

Agent gallery showing multiple demo agents

Teams shell

The customer forgets it's a prototype

Pixel-accurate Microsoft Teams chrome — top bar, icon rail, sidebar with recent chats, chat area with real streaming output. The customer types their own questions and gets real LLM responses, not scripted answers.

The About tab shows what the real production build would take — purpose, integrations, security, and escalation paths. The Shared tab surfaces files, links, and documents associated with the agent.

Fluent UI v9 streaming tokens structured output blocks

Agent creation

Three ways to build — the fastest takes under 5 minutes

Paste from LLM: copy the generation prompt, describe the use case in any LLM, paste the response. Agent is live.

Form Builder: six-step in-browser configurator for identity, persona, avatar, language, prompts, and content stubs — no code editor needed.

CLI wizard: npm run new-agent — answer questions in the terminal, Claude writes the agent folder directly.

no code required saves to disk zip export

What comes out the other end

Demo to proposal input,
in one session

The tool doesn't just run demos. It closes the loop from first conversation to ROM/AEP scoping.

💬

Live, interactive agent demo

The customer interacts directly with a Teams-style agent that knows their systems, handles their use cases, and responds in real time. Not a walkthrough — a conversation.

📋

Client questionnaire

Structured discovery notes for the customer — a professional follow-up artefact that shows you understand their problem space. Exportable after the session.

📐

FDE Design Report

21-section internal framework covering complexity, governance, integration, data, security, operating model, ROI, and adoption maturity. Directly feeds the SA's ROM/AEP scoping session.

📁

Agent folder (reusable)

Every agent you build is a portable folder — drop it into any repo clone, share it with another FDE, or iterate on it for the next engagement with the same client.

Who this is for

Built for three roles
in one session

⚡ FDEs

No code needed to build an agent
Personalise to any client in minutes
Switch between Anthropic, Azure AI, OpenAI
Reuse and iterate agent folders across engagements
Capture discovery data without a separate tool

🏗️ Solution Architects

About tab shows build complexity and platform choice
FDE Design Report covers all 21 scoping areas
Feeds directly into ROM/AEP without a separate briefing
Integration scope and data architecture captured automatically

📈 Sellers

Customer touches a working agent, not a deck
Objections answered in the demo, not after it
Client questionnaire is a ready-to-share follow-up
Consistent demo quality across the practice

Quick start

Running in under 5 minutes

Node 18+, one API key, and this repo. That's the full dependency list.

# Clone and install
git clone https://github.com/ava-innersource/fde-demo-worlds.git
cd fde-demo-worlds
npm install

# Drop your key in .env
cp .env.example .env
VITE_ANTHROPIC_API_KEY=sk-ant-...   # or Azure AI / OpenAI

# Start the dev server
npm run dev

# Open http://localhost:5173 — HR agent is live, ready to demo
# Build your first agent: click ✦ logo → Paste from LLM

Need to demo on a client's Azure AI contract? Switch providers in /setup without restarting. The agent configs are provider-agnostic.

A live AI agent demo,in under 20 minutes.

The demo you could showvs the deck you're showing

From cold startto customer conversation