DataBard

Weekly data health briefings your team actually consumes.

Data teams produce findings nobody reads. The exec doesn't open Looker. The PM doesn't check dbt tests. The data engineer's Slack message about "the payments table has been stale for 3 days" gets buried. The insight exists but it doesn't land.

DataBard closes that gap. Every Monday morning, your team gets a fresh audio briefing on your data health — health scores, what changed, what to fix. No dashboard to check. No report to read. Just press play.

▶ Listen to a demo episode — no signup required

---

The Problem

Data health monitoring tools tell you what's broken. They don't tell the people who need to know. The result:

• Data leads spend hours building reports nobody opens. Dashboards have 47 rows of test results. Execs want a 1-line summary.
• Issues surface too late. By the time someone notices the payments table is stale, 8 downstream dashboards are already wrong.
• Audit trails live in tools auditors can't access. Compliance wants proof of data quality history. Engineering has it in dbt logs nobody reads.

The Solution

DataBard is one analysis engine with outputs people actually consume. The engine ingests metadata from any source, computes health scores, critical-table rankings, coverage gaps, and trend narratives. Every output — audio briefing, dashboard, alert, audit record — renders from that same analysis.

The wedge: the weekly digest

The core product is the scheduled weekly digest. Every Monday, your team receives:

1. A 2-minute executive briefing (audio) — top 3 issues, what changed, recommended actions
2. A dashboard with health scores, trend narratives, and drill-down
3. An alert if health dropped below your threshold since last week

The audio is the differentiator. It's not a gimmick — it's a forcing function for synthesis. You can't read 47 rows aloud. You have to say "your payments table is stale, 8 dashboards are wrong, and here's what to do." That synthesis is the value.

The dashboard is the hero

After every analysis, you land on the dashboard — not the audio player. Health scores, critical tables, trend narratives, and "What changed this week" are front and center. The audio is a button on the dashboard: ▶ Listen to this analysis.

This makes the product feel like an analyst that also talks — not a podcast that also analyzes.

Architecture

┌─────────────────────────────────────────────┐
│              Data Sources                    │
│  OpenMetadata · dbt · The Graph · Dune       │
│  Coral (50+ sources via SQL)                 │
└───────────────────┬─────────────────────────┘
                    │
                    ▼
          ┌──────────────────┐
          │  Analysis Engine  │
          │  Health score     │
          │  Critical tables  │
          │  Trend narratives │
          │  PII / governance │
          └────────┬─────────┘
                   │
       ┌───────────┼───────────┐
       ▼           ▼           ▼
  ┌─────────┐ ┌─────────┐ ┌─────────┐
  │Dashboard│ │  Audio  │ │ Alerts  │
  │ (hero)  │ │Briefing │ │Webhook/ │
  │Trends   │ │2-15 min │ │Slack    │
  │Drill-down│ │Exec/Full│ │Email    │
  └─────────┘ └─────────┘ └─────────┘
       │           │           │
       └─────┬─────┘           │
             ▼                 ▼
      ┌────────────┐    ┌────────────┐
      │  Solana    │    │  Weekly    │
      │Attestation │    │  Digest    │
      │(Onchain    │    │  (Pro)     │
      │ persona)   │    │            │
      └────────────┘    └────────────┘

Pipeline: Connect source → Metadata fetch → Schema analysis (health score, critical tables, trends) → Dashboard (hero) → Audio briefing (button on dashboard) → Optional: schedule weekly, alert on drops, attest on-chain

Output Formats

Format	Length	Use case
Executive briefing	2 min	Monday morning — top 3 issues + actions
Full analysis	10-15 min	Deep dive — two AI hosts discuss every finding
Dashboard	—	Health scores, trends, drill-down (always available)
Alert	—	Slack/webhook when health drops below threshold
Onchain attestation	—	Tamper-evident audit trail (Onchain persona)

Features

• Dashboard-first flow — after analysis, land on the dashboard with health scores and trend narratives. Audio is a button, not the destination.
• Trend narratives — "What changed this week" section computes week-over-week diffs and explains them in plain English: "Health dropped 8 points because test coverage fell in the payments schema after the Friday deploy."
• Executive summary format — 2-minute briefing with top 3 issues, what changed, and recommended actions. For busy people who want the bottom line.
• Two-voice AI podcast — Alex (advocate) and Morgan (auditor) via ElevenLabs. Full analysis mode for deep dives.
• Alerts — health threshold monitoring with Slack/webhook notifications. Decoupled from wallet — works with email only.
• Scheduled weekly digests — Pro tier. Set up a Monday morning briefing for your team. RSS + email delivery.
• Onchain attestation — Solana Memo Program. Tamper-evident audit trail for data quality history. Onchain persona only.
• Coral integration — 50+ data sources via SQL. Cross-source joins, no ETL, no data warehouse.
• Interactive drill-down — click any segment in the audio player to see columns, tests, lineage, tags.
• Labs — experimental features including Data Anthems (data-driven songs). Accessible at /labs.

Quick Start

git clone https://github.com/thisyearnofear/databard.git
cd databard
npm install

cp .env.example .env
# Add your ELEVENLABS_API_KEY (Starter plan or higher recommended)
# Add OPENAI_API_KEY for LLM scripts (optional)
npm run dev

ElevenLabs Setup:

1. Sign up at elevenlabs.io
2. Upgrade to Starter plan ($5/month) for API access
3. Get your API key from Profile → API Keys
4. Add to .env: ELEVENLABS_API_KEY=sk_your_key_here

Open localhost:3000 → Connect → Dashboard with health scores → Listen to briefing.

Tech Stack

Layer	Technology
Web UI	Next.js 16, React 19, Tailwind CSS 4
Data sources	OpenMetadata REST API, dbt manifest, The Graph, Dune Analytics, Coral (SQL)
AI scripts	OpenAI-compatible API (GPT-4o-mini default; Azure OpenAI drop-in — docs/AZURE.md)
Audio	ElevenLabs TTS (two voices) + Sound Effects
Caching	File-backed with TTL (no external dependencies)
Payments	Stripe Checkout, Palm USD (Solana SPL stablecoin)
Onchain	Solana Memo Program + PDA registry (Onchain persona)

Engagement Loops

ATTENTION
  ├─ Social post: "AI roasted my database" (shareable clip)
  ├─ Colleague forwards Slack link to shared episode
  └─ Blog post: "We replaced our weekly data report with a podcast"
       │
       ▼
CONVERSION (first "wow" moment)
  ├─ Demo: hears AI hosts discuss sample schema (zero friction)
  ├─ Connects own data → sees health score → hears personal analysis
  └─ "Roast my data" emotional trigger → shares result
       │
       ▼
RETENTION (habit formation)
  ├─ Sets up weekly digest (Monday morning briefing)
  ├─ Alert fires when health drops → comes back to dashboard
  ├─ Trend narrative: "what changed?" → curiosity pull
  └─ Team forwards digest in Slack → social accountability
       │
       ▼
VIRAL (each retention cycle creates new attention)
  ├─ Shared episode link in Slack → 5-20 colleagues click
  ├─ Health score badge in README / team page
  ├─ "Share this moment" clip → social media
  └─ Weekly digest email forwarded to stakeholders
       │
       └─── (loops back to ATTENTION)

See docs/GTM.md for the full go-to-market strategy, viral hooks, and user interview plan.

Roadmap

Done

• Dashboard-first flow (land on dashboard after analysis, not player)
• Executive summary format (2-minute briefing)
• Trend narratives ("What changed this week")
• Format picker for all sources (not just Coral)
• Anthem moved to /labs (experimental, not in main flow)
• Alerts page with email-based subscriptions (no wallet required)
• Scheduled weekly digests (Pro tier)
• Onchain attestation (Onchain persona)
• Coral integration (50+ sources via SQL)
• Two-voice AI podcast with ElevenLabs
• Health analytics dashboard with sparklines and trend badges
• Solana on-chain audit trail, leaderboard, team history

Next — Viral & Retention

• "Get this for your data" CTA on shared episode pages
• "Want this every Monday?" one-click schedule from dashboard
• "Share this moment" clip feature (15-second audio highlight)
• Email delivery for scheduled digests
• "Roast my data" landing page variant
• Health score badge (embeddable SVG for README/team page)
• Team email recipients for scheduled digests

Future

• Azure migration — inference on Azure OpenAI, hosting on Container Apps (docs/AZURE.md)
• Microsoft Purview Tier-1 adapter (docs/PURVIEW_ADAPTER.md)
• Custom voice personalities
• Benchmarking — "your health score vs. teams your size"

Solana Integration

DataBard uses Solana as a verifiable audit trail for the Onchain persona. Every health report can be attested on-chain — a tamper-evident record your team and auditors can verify.

Feature	Description
Onchain attestation	Every episode mint writes a memo + PDA record: schema, health score, timestamp, episode ID, wallet
Health alert subscriptions	Register wallet + schema + threshold + webhook; fires when health drops
Public leaderboard	Ranked protocols by health score + trend — /leaderboard
Team history	Cross-wallet mint history per schema — shared ground truth for post-mortems
Palm USD payments	Pay for Pro with Palm USD, a non-freezable Solana stablecoin
SNS .sol identity	Wallet address resolved to .sol domain on connect

See docs/PALM_USD_INTEGRATION.md for Palm USD setup.

ElevenLabs Integration

Host	Personality	Voice
Alex	Enthusiastic data advocate	George (JBFqnCBsd6RMkjVDRZzb)
Morgan	Skeptical quality auditor	Charlotte (XB0fDUnXU5powFXDhCwa)

Both voices use context stitching (previous_text / next_text) for natural prosody across segment boundaries. Transitions use ElevenLabs Sound Effects API (cached 30 days).

Setup:

# ElevenLabs Starter plan ($5/month) required for API access
ELEVENLABS_API_KEY=sk_your_key_here

TestSprite Verification Loop

DataBard is submitted to TestSprite Season 3 — "CLI Launch & Loop Engineering."

The loop defends economic invariants — properties that must hold or the market silently breaks:

1. Digest reseller must earn positive margin on every settled deal
2. Cascade wins Quality briefs, Newsroom wins Freshness — persona-focus fit
3. Escrow state machine rejects invalid transitions (no release before commit)
4. Release cascade settles all sub-escrows (Newsroom must get paid)

Each invariant is a Python + requests test in tests/testsprite/. The loop (scripts/loop/loop.mjs) uploads them to TestSprite Cloud, runs them against the live URL, and on failure feeds the failure bundle to the coding agent to propose + apply + commit a minimal patch. Re-runs until green or hits the 4-iteration cap.

CI/CD: The TestSprite checker is wired into GitHub Actions (.github/workflows/testsprite.yml) — every PR reruns the invariants and fails the build if something breaks. (+5 Innovation)

See LOOP.md for the full iteration audit trail — every run, every fix, every commit SHA.

License

MIT