Skip to content

NorthStar AI

An end-to-end AI workflow where GPT-5 infers KPIs, Codex generates SQL/Python pipelines, and the Responses API orchestrates reporting-automating product analytics from PRD to polished report.

Code Video Cerebral Valley

Core trick

Has an LLM generate SQL and Python for product KPIs, then runs that code in DuckDB to produce reports.

Tags

productivityagents

Description

An end-to-end AI workflow where GPT-5 infers KPIs, Codex generates SQL/Python pipelines, and the Responses API orchestrates reporting-automating product analytics from PRD to polished report.
vatsalajha/GPT5-Hackathon★ 02 forksPythonCreated Sep 27, 2025 · last push Sep 28, 2025

README excerpt

# GPT5-Hackathon
# Shadow Product Manager Agent

Comprehensive agentic workflow that mirrors a product manager's analytics loop. The Streamlit UI ingests files or Google Drive links, a suite of OpenAI-powered planners proposes KPIs and commentary, and DuckDB runs generated SQL/Python to ship stakeholder-ready reports. This README documents the full system that lives on the `gpt5-model-suggests-objective` branch.

---

## Quick Start

1. **Python environment** – create and activate a Python 3.10+ virtualenv.
2. **Install dependencies**
   ```bash
   pip install -r Workflow/requirements.txt
   ```
3. **Configure secrets**
   ```bash
   export OPENAI_API_KEY="sk-..."
   ```
   Optional: copy `.env.example` to `.env` and place the key there; Streamlit reads it on launch.
4. **Run the Streamlit UI**
   ```bash
   cd Workflow
   streamlit run streamlit_app.py
   ```
5. Paste a business objective, upload local datasets (CSV, TSV, XLSX, Parquet, JSON, PDF, TXT, PNG/JPG, or archives), or provide remote Google Drive links. Click **Generate shadow PM report**.

Outputs include KPI tables/visuals, an HTML report in `Workflow/reports/`, and a stakeholder email draft.

---

## System Overview

```
Streamlit UI (streamlit_app.py)
 ├─ Objective suggestion from uploaded docs/images
 ├─ Remote download + cache layer (agent.utils.fetch_remote_uploads)
 └─ ShadowPMOrchestrator
      ├─ DataIngestor → Pandas profiles, PDF text, image serialisation
      ├─ KpiGenerator → LLM proposes KPIs + SQL/P

Team