From Dashboards to Dialogue: How We Embedded Conversational AI in Our BI Platform

Real-Time Answers. Embedded Intelligence. No SQL Required.

The Reality of Traditional BI

We’ve worked with clients across industries—SaaS, travel, healthcare—and they all have one thing in common: mountains of dashboards.

And yet, when leadership or business teams need a specific answer, it goes like this:

1. Raise a data request ticket
2. Wait for data engineering to write the query
3. Review a dashboard that partially answers the question

That’s days lost.
The insight gap costs more than time—it’s missed decisions, lost opportunities, and frustration.

We knew there had to be a better way.

The Idea: Let Users Talk to the Dashboard

We envisioned a dashboard where you didn’t need to filter, click, or drill-down endlessly. Instead, you could simply ask:

• “Show weekly growth in signups for the new onboarding flow.”
• “What’s the revenue contribution from tier-1 cities last quarter?”
• “Compare churn between mobile and web users for the past 6 months.”

That vision led us to build an integrated Conversational BI system using our MCP infrastructure.

How We Implemented It

1. Data Warehouse Foundation with PostgreSQL

We started by cleaning and modeling data into PostgreSQL. Our ETL layer handled unification across product analytics, sales, and CRM systems. Daily and hourly syncs were orchestrated via Airflow to ensure data freshness.

2. Interactive Dashboards with Apache Superset

Apache Superset was deployed independently within our analytics environment on a dedicated VM. Superset connected to our PostgreSQL data warehouse and provided a rich interface for building charts, KPIs, and role-based dashboards.

3. Conversational Layer with Embedded LLM

Here’s where the transformation happened:

• We fine-tuned the open-source LLaMA 2 model using a curated dataset of analytics queries specific to SaaS and travel domains.
• Built a custom lightweight chatbot frontend in React and embedded it within Superset using iframe integration.
• The chatbot converted natural language queries → into SQL → queried PostgreSQL → and returned chart responses or tabular summaries inline.

4. Secure by Design: Full On-Prem Deployment

No external cloud APIs. No metadata leakage.
All components—Superset, PostgreSQL, LLM, and the chatbot interface—ran securely on our self-hosted MCP server under enterprise-grade authentication and access policies.

Real Use Case: Daily Ops Brief for a SaaS Team

Every morning, product and sales teams needed to review KPIs like:

• Signups by channel
• Demo conversion rate
• Support tickets opened vs. resolved

We connected Airflow pipelines, PostgreSQL metrics tables, and Superset.

Now, instead of flipping through 7 dashboards, our product lead types:

“Summarize yesterday’s conversions and top 3 customer issues.”

Within seconds, the chatbot:

• Pulled signup and conversion data from PostgreSQL
• Extracted top Zendesk issues by ticket count
• Displayed visual charts + a summary sentence

Team got their daily ops briefing in <10 seconds. Zero SQL. Zero wait.

Key Benefits We Observed

• Speed: Follow-up questions answered instantly
• Accessibility: Business users didn’t need technical skills
• Productivity: Data teams focused on deep analysis, not daily queries
• Security: Data never left the internal network

Looking Ahead

We’re continuously improving:

• Adding context memory for multi-turn queries
• Supporting drill-down commands via chat (e.g., “break down by device type”)
• Including query change tracking and audit logs

Final Thoughts

Dashboards helped us visualize the past.

With conversational BI, we empower teams to interrogate the present and simulate the future—securely, instantly, and intuitively.

If you’re exploring ways to make data actually usable across your organization, we’re happy to share our learnings.

Please leave this field empty

Stay in the Loop with HashStudioz Blog

Check your inbox or spam folder to confirm your subscription.