Most people interact with data through dashboards — static views built in advance for questions someone already thought to ask.
AskMyData flips that.
Type any question in plain English. The app understands what you're asking, writes the SQL, runs it against a real database, and explains what it found — in seconds. No SQL knowledge required. No pre-built reports. No waiting for an analyst.
"Which department had the most critical incidents last quarter?" "What's the average time to resolve high severity incidents?" "Show me all ransomware incidents in the Finance department."
Any question. Instant answer. Plain English explanation.
┌─────────────────────────────────────────────────────────────────┐
│ USER INTERFACE │
│ Streamlit Web App · Python · Light Theme │
└────────────────────────┬────────────────────────────────────────┘
│
│ Natural language question
▼
┌─────────────────────────────────────────────────────────────────┐
│ LLM LAYER │
│ Large Language Model · Natural Language Understanding │
│ │
│ Input: User question + database schema │
│ Output: Valid SQLite query │
│ │
│ Prompt engineering ensures: │
│ → SELECT only (no destructive queries) │
│ → Exact column name matching │
│ → SQLite-compatible syntax │
│ → Context-aware query construction │
└────────────────────────┬────────────────────────────────────────┘
│
│ Generated SQL query
▼
┌─────────────────────────────────────────────────────────────────┐
│ DATA LAYER │
│ SQLite · In-Memory Database · Pandas │
│ │
│ → Excel dataset loaded into SQLite on app start │
│ → SQL executed against in-memory database │
│ → Results returned as structured dataframe │
│ → Auto chart generated for numeric results │
└────────────────────────┬────────────────────────────────────────┘
│
│ Query results
▼
┌─────────────────────────────────────────────────────────────────┐
│ INSIGHT LAYER │
│ LLM · Analytical Reasoning · Plain English Output │
│ │
│ Input: User question + SQL + results │
│ Output: 2-3 sentence executive-level insight │
│ │
│ Explains: What the data means · Patterns · Recommendations │
└─────────────────────────────────────────────────────────────────┘
│
│ Answer + insight + chart
▼
👤 User sees results
This isn't a keyword matcher or a rule-based query builder. It uses a Large Language Model (LLM) for two distinct tasks:
The LLM receives the user's question alongside the full database schema — column names, data types, and valid values. It reasons about what the user is asking, maps it to the correct columns and filters, and generates a syntactically valid SQLite query.
Prompt engineering handles:
- Safety — only SELECT queries are permitted, never destructive operations
- Accuracy — exact column names enforced, correct data types respected
- Context — date formats, text comparisons, and aggregation logic handled correctly
- Edge cases — ambiguous questions are resolved using schema context
Once the query runs and results are returned, the LLM receives the original question, the SQL, and the result set. It synthesises these into a 2-3 sentence analytical insight written for a non-technical audience — specific numbers, pattern observations, and one actionable recommendation.
This is the difference between data and insight.
200 cybersecurity incidents · 18 dimensions · Jan 2024 – Apr 2025
| Dimension | Values |
|---|---|
| Incident Type | Phishing, Malware, Ransomware, Unauthorized Access, DDoS, Insider Threat, Data Exfiltration, Credential Compromise, Vulnerability Exploit, Social Engineering |
| Severity | Critical / High / Medium / Low |
| Status | Open / In Progress / Resolved / Closed |
| Department | IT, Finance, HR, Operations, Legal, Marketing, Engineering, Compliance |
| Region | North America, Asia Pacific, Europe, Latin America |
| Attack Vector | Email, Network, Web Application, Insider, Physical, Third Party, Cloud |
| Control Gap | MFA Not Enforced, Patch Management Delay, Excessive Privileges, and more |
| Financial Impact | CAD dollar value per incident |
| SLA Performance | Met vs Breached by severity tier |
| Root Cause | Human Error, System Vulnerability, Process Gap, Third Party, Unknown |
🔍 Natural language query Type any business question in plain English — no SQL knowledge needed
📊 Embedded live dashboard Full interactive Power BI dashboard embedded directly in the app — explore visually before querying
💡 AI-generated insights Every query result comes with a plain English analytical explanation written for non-technical stakeholders
⚡ 10 example prompts One-click example questions to explore the dataset instantly
📈 Auto chart generation Numeric results automatically render as bar charts for quick visual interpretation
🔒 Secure by design API key stored in encrypted secrets vault
git clone https://github.com/Swarleen/askmydata
cd askmydata
pip install -r requirements.txt
streamlit run app.pyAdd your API key to .streamlit/secrets.toml:
ANTHROPIC_API_KEY = "your-api-key-here"Open http://localhost:8501
- Fork this repo
- Go to share.streamlit.io → connect GitHub → select repo
Key concepts applied:
- Large Language Model (LLM) integration for natural language understanding
- Prompt engineering for safe, accurate SQL generation
- In-memory relational database query execution
- LLM-powered result interpretation and insight generation
- Secure secret management for API credentials
- Streamlit session state management for interactive UX
Swarleen Bhamra — Business Analyst & Data Analyst based in Toronto, ON.
I build data solutions that sit at the intersection of analytics, AI, and business intelligence — from Power BI dashboards and audit analytics tools to AI-powered applications that make data accessible to everyone.
| 🌐 Portfolio | swarleen.com |
| linkedin.com/in/swarleenbhamra | |
| 📊 Power BI Dashboards | github.com/Swarleen/PowerBI-Portfolio |
| 🐙 GitHub | github.com/Swarleen |
💼 Open to Business Analyst, Data Analyst, and Technology Analytics roles in Toronto. If this project resonates with what your team is building — let's connect.
