DataPilot

Drop a CSV, ask a question, get charts + insights in seconds.

AI data analyst powered by GPT-4o code generation. No SQL, no Excel, no code — just ask in plain language and get answers with visualizations.

Try it live

---

How it works

CSV upload → Schema Detection → GPT-4o Codegen (pandas) → Sandbox Execution → Answer + Chart
                                                                    ↓ (optional)
                                                    GPT-4o-mini → Proactive Insights + Chart

1. Your CSV is parsed and its schema extracted (columns, types, stats, sample values)
2. GPT-4o generates pandas code to answer your question (JSON mode)
3. The code runs in a sandboxed environment — if it fails, one automatic retry with error feedback
4. You get a text answer + optional chart (dark-themed matplotlib/seaborn)
5. Optionally, GPT-4o-mini discovers 3-5 proactive insights with computed evidence

Evaluation results

Round	Model	Accuracy	Hallucinations	Decision
R1	GPT-4o-mini	55% (11/20)	1	NO-GO
R2	GPT-4o + anti-hallucination prompt	87.5% (17.5/20)	0	CONDITIONAL GO

Evaluated on 20 golden questions (aggregation, trend, comparison, multi-step, adversarial). Median latency: 2.1s. Full eval: docs/EVAL-REPORT.md

Key decision: model upgrade from GPT-4o-mini to GPT-4o was driven entirely by eval results — not intuition.

Tech stack

Component	Technology
LLM (main)	GPT-4o (OpenAI) — code generation
LLM (insights)	GPT-4o-mini (OpenAI) — proactive insights
Execution	pandas + matplotlib/seaborn in sandboxed exec()
Backend	FastAPI (Python)
Frontend	React + Tailwind (Lovable)
Hosting	Render ($7/mo)

Local setup

# Clone and setup
git clone https://github.com/Mehdibargach/datapilot.git
cd datapilot
python3 -m venv venv && source venv/bin/activate
pip install -r requirements.txt

# Add API keys
cp .env.example .env
# Edit .env with your OPENAI_API_KEY

# Run
uvicorn api:app --host 0.0.0.0 --port 8000

API endpoints:

• GET /health — health check
• GET /datasets — list demo datasets (superstore, saas, marketing)
• POST /analyze — ask a question about your CSV

Project structure

api.py              ← FastAPI backend (3 endpoints)
agent.py            ← Pipeline orchestrator (analyze + insights)
codegen.py          ← GPT-4o code generation + meta-question detection
sandbox.py          ← Sandboxed pandas code execution
schema.py           ← CSV schema detection (multi-encoding)
insights.py         ← GPT-4o-mini proactive insight discovery
data/               ← Demo datasets (superstore, saas, marketing)
docs/
  EVAL-REPORT.md    ← Full eval with golden dataset (20 questions)
  ADR.md            ← Architecture Decision Records

Built by

Mehdi Bargach — Builder PM · AI Products

LinkedIn