BangHeri AI Data Analyst — AI-Powered CSV Analysis with Secure BYOK Architecture

Upload CSV → ask questions → AI analyzes, charts, exports PDF — secure BYOK architecture.

BangHeri AI Data Analyst is a production-ready web application that lets business users and analysts upload CSV datasets, ask natural-language questions, and receive AI-generated analysis with charts and downloadable PDF reports — all without writing code.

Problem Solved

Data analysts and business users need quick, code-free insights from CSV data. Existing tools either require coding (Python/Pandas), leak API keys to servers, or lack exportable reports. BangHeri AI Data Analyst solves this with a zero-server-key architecture: users bring their own Gemini API key (held only in browser memory), upload CSVs inline (≤1 MB), and receive AI-generated analysis with charts and PDF reports — all running in a managed sandbox.

Tech Stack

Layer	Technology
Frontend	React 19, Vite, TypeScript, Tailwind CSS v4, jsPDF
Backend	Express (tsx), SSE streaming, Google Interactions API proxy
AI Agent	Gemini antigravity-preview-05-2026 + code_execution tool
Sandbox	Google-managed Linux (pandas, numpy, matplotlib pre-installed)
PDF Export	jsPDF 4.2.1 (deterministic A4, width-aware title ellipsis)
Automation Ready	n8n + GOWA webhook integration for WhatsApp bot

Key Features

• BYOK (Bring Your Own Key): API key never touches server — sent per-request via x-gemini-api-key header, held only in browser memory.
• Inline CSV upload: ≤1 MB, no cloud storage dependency.
• Natural language analysis: Ask any business question; agent profiles, computes, visualizes.
• Chart generation: Matplotlib charts rendered in sandbox, extracted via tar snapshot, served statically.
• Deterministic PDF export: A4, width-aware KPI titles, no clipping/overlap, verified via controlled test harness (CP8.1–CP8.3).
• Follow-up conversations: Reuses warm sandbox (environmentId) for multi-turn analysis.
• Fail-closed design: Missing key → 400; no secret in env; quota bypass in dev.

Production Evidence

Checkpoint	Result
Health check (/api/health)	PASS HTTP 200 {"status":"ok"}
Upload endpoint (/api/upload)	PASS HTTP 200, file echoed, 1 MB cap
Analyze gate (missing key)	PASS HTTP 400 fail-closed, no provider call
Quota endpoint (/api/quota)	PASS HTTP 200 {"used":0,"limit":999999}
Deploy commit	267c814b — controlled push, manual Render deploy
Tag v1.0.0	Immutable at 7ad11fd (pre-patch baseline)
Production PDF smoke test	PASS Verified via production deployment

Links

• Live Demo: https://bangheri-ai-data-analyst.onrender.com
• GitHub Repo: https://github.com/herirahmansyah/bangheri-ai-data-analyst

← Back to Home