Build AI Assistant in 10 Minutes | Misar Blog | Assisters

Here’s a clean, technical walkthrough in Markdown:

## Introduction

Building your first AI assistant no longer requires a PhD in machine learning or weeks of coding. Modern “assistant-as-a-service” platforms like **Assisters** let you scaffold, train, and deploy a conversational agent in minutes. In this guide, we’ll create a simple “Weather Buddy” that answers questions about the weather in any city. By the end, you’ll have a working endpoint that you can embed in a Slack workspace, a website, or even a Raspberry Pi.

---

## What You’ll Need

- An [Assisters](https://assisters.ai) account (free tier available)
- A modern browser (Chrome, Firefox, Edge)
- A text editor (VS Code, Sublime, or even Notepad)
- Optional: a Slack workspace for live testing

---

## Step 1: Create a New Assistant

1. Log in to the Assisters dashboard and click **New Assistant**.
2. Give it a name: `Weather Buddy`.
3. Choose **Template: Q&A Bot** (we’ll swap the default knowledge base later).
4. Click **Create Assistant**.

You’ll land on the **Build** tab, where the assistant skeleton is already live at a temporary URL like `https://assisters.ai/bot/weather-buddy-abc123`.

---

## Step 2: Swap the Knowledge Base for Real Data

Out of the box, the template answers generic questions. We need live weather data.

### Option A – Quick Weather API Wiring (Recommended)

1. Go to **Integrations** → **Add Integration**.
2. Search for **OpenWeatherMap**.
3. Paste your API key (get one at [openweathermap.org](https://openweathermap.org/api)).
4. In the **Prompt Template** field, paste:

handlebars {{#each messages}} {{#if (contains text "weather")}} Current weather in {{extractCity text}} is {{callOpenWeatherMap "weather" city=extractCity text}}°C. {{/if}} {{/each}}

- `extractCity` and `callOpenWeatherMap` are built-in helper functions that parse the user’s sentence and call the API.

### Option B – Pure LLM Prompting (No API Key)

If you don’t want to add an external API, you can hard-code a few cities:

1. Go to **Train** → **Add Example**.
2. Type:

What's the weather in Chicago?

3. Add the reply:

Chicago: 22°C, partly cloudy.

Repeat for London, Tokyo, etc. (≈10 examples trains the model well.)

---

## Step 3: Test the Assistant Live

1. Open the **Test** pane on the right.
2. In the chat box, type:

What's the weather in Berlin?

You should see either:

- **Option A**: `Berlin: 18°C, light rain.` (fetched live)
- **Option B**: `Berlin: 18°C, light rain.` (pre-canned)

If you see “I don’t know,” double-check your prompt template or example list.

---

## Step 4: Add Personality & Safety

1. **Personality**: In **Build** → **Tone**, set:

You are a friendly weather expert who answers in 1–2 sentences.

2. **Safety**: Under **Moderation**, enable:

- Hate-speech filter
- PII redaction (emails, phone numbers)

3. **Fallback**: In **Build** → **Default Reply**, set:

Sorry, I can only talk about the weather. Try “What’s the weather in Paris?”

---

## Step 5: Publish to a Public URL or Slack

### Public Endpoint

1. Go to **Deploy** → **Public URL**.
2. Click **Enable**.
3. Copy the HTTPS endpoint (`https://assisters.ai/bot/weather-buddy-abc123`). You can now:
   - Embed it in a `<iframe>` on a webpage.
   - Call it from Postman with `POST /message { "text": "weather in Sydney" }`.

### Slack Channel

1. Go to **Deploy** → **Slack**.
2. Click **Add to Slack** and pick your workspace.
3. Choose either:
   - **Bot Token**: posts as @WeatherBuddy
   - **Incoming Webhook**: posts as your username
4. In Slack, invite `@WeatherBuddy` to any channel and ask:

/invite @WeatherBuddy

Then test:

/weather-buddy what's the weather in Madrid?

---

## Step 6: Saving & Versioning

1. Click **Save Draft** after every change.
2. When satisfied, click **Publish** in the top-right corner.
3. Assisters auto-creates a semantic version (`v1.0.0`). Any new edits go to `v1.1.0-dev`, which you can promote to production later.

---

## Step 7: Monitoring & Analytics

After a few hours of traffic:

1. Go to **Analytics** → **Conversation Logs**.
2. Filter for `intent:weather` to see how many users asked about the weather.
3. Check **Accuracy** to see if the assistant answered correctly.
4. If accuracy < 95%, add more examples in **Train**.

---

## Behind the Scenes: How It Works

| Component | Technology |
|-----------|------------|
| Intent parsing | spaCy + custom regex |
| Slot filling | Handlebars helpers (`extractCity`) |
| External API | OpenWeatherMap (REST) |
| LLM backend | Assisters proprietary transformer (≈ 2B params) |
| Hosting | AWS Lambda + CloudFront edge cache |

The entire pipeline runs in < 300 ms P99 latency worldwide.

---

## Going Further

- **Multi-language**: Add Spanish examples in **Train** → **Add Language**.
- **Voice**: Enable **Deploy** → **Voice** to accept voice messages.
- **Custom actions**: Call a custom Lambda via **Integrations** → **Webhook**.
- **White-label**: Upload a logo in **Branding** and replace the public URL with your own domain (`weather.yourbrand.com`).

---

## Final Thoughts

In ten minutes you’ve built an AI assistant that answers weather queries, speaks to Slack users, and deploys globally behind a single HTTPS endpoint. The real power isn’t in the weather—it’s in the template-driven approach that lets anyone wire together LLMs, APIs, and UIs without touching a terminal. Once you’re comfortable, try swapping the weather API for a stock ticker, a calendar client, or even a game leaderboard. The only limit is your imagination (and the free tier’s 10k messages/month).