LinguaBot.

The problem.

The South Asian small-business web has a multilingual gap. Most chat widgets ship in English only, or require manual language switching. Customers in Dhaka, Karachi, or Mumbai land on an English-only chat and bounce. The off-the-shelf options (Intercom, Drift) are fifty times the price a 3-person consultancy can stomach.

What I built.

• Auto language detect. The widget reads the visitor's first message and switches UI + responses to EN / BN / HI / UR on the fly.
• One script tag. Drop in <script src="https://linguabot.app/w/xyz.js">. No npm, no build step, no React peer dependency.
• Free tier forever. Up to 100 conversations / month at zero cost. Paid plan kicks in only when you've already validated the value.
• Self-hostable. Open core. You can run the whole stack on your own Fly machine and pay $5/mo instead of $50.
• FastAPI + React dashboard. For tenant config, prompt templates, language fallbacks, and analytics.

How I shipped it.

1. Widget first. Vanilla JS + Shadow DOM, 12kb gzipped. Designed to never break a host site's CSS.
2. Dashboard last. Tenants got CLI config first — only built the React admin once five paying users asked for it.
3. Multi-tenant from day one. Per-tenant API keys, prompt overrides, language mix. PostgreSQL row-level security.
4. Stripe + webhook reconciliation. One job, runs nightly, never gets confused about a plan downgrade mid-month.
5. Fly.io deploy. $5/mo to start. Scales horizontally without me having to think about Kubernetes.

// One script tag. Auto-detects language. Free tier forever.
<script
  src="https://linguabot.app/w/xyz123.js"
  data-position="bottom-right"
  data-lang="auto"
></script>

async def detect_lang(text: str) -> Lang:
    # Bengali script range
    if any('ঀ' <= c <= '৿' for c in text): return Lang.BN
    # Devanagari (Hindi)
    if any('ऀ' <= c <= 'ॿ' for c in text): return Lang.HI
    # Arabic / Urdu
    if any('؀' <= c <= 'ۿ' for c in text): return Lang.UR
    return Lang.EN  # default fallback

Outcomes.