03 — Setting up the environment

Duration: 15 min (≥ 30 min the first time if you download the 4.9 GB model) Prerequisites: a terminal and an internet connection.

Overview

Component	What it does	Size
Python 3.9+	Runtime	~50 MB
csv-llm-shared	Ingestion, normalization, DB, dashboards	~200 KB
csv-llm-ollama	The Streamlit app	~100 KB
Ollama	Local LLM server	~600 MB
llama3.1:8b	The tool-calling model	~4.9 GB

You’ll clone two repos side-by-side: the app and the shared library it depends on.

1. Python and the two repos

The project lives in two sibling repositories. Clone them into one folder of your choice (here we use project-1/).

Windows (PowerShell)

python --version
# expected: Python 3.9.0 or higher

mkdir project-1
cd project-1

# Shared library (mandatory: ingest, normalize, db, dashboards, data/)
git clone https://github.com/inskillflow/csv-llm-shared.git

# The Ollama app
git clone https://github.com/inskillflow/csv-llm-ollama.git

macOS / Linux

python3 --version
mkdir project-1 && cd project-1
git clone https://github.com/inskillflow/csv-llm-shared.git
git clone https://github.com/inskillflow/csv-llm-ollama.git

After cloning you should have:

project-1/
├── csv-llm-shared/   ← shared Python module + data fallback
└── csv-llm-ollama/   ← Streamlit app + local Ollama + data/

Virtual environment (optional but clean)

python -m venv .venv
.\.venv\Scripts\Activate.ps1   # PowerShell
# or: source .venv/bin/activate  (bash/zsh)

2. Install Python dependencies

Two requirements.txt, one per repo. Run from the parent folder (project-1/):

pip install -r csv-llm-shared/requirements.txt    # pandas, plotly, pyyaml
pip install -r csv-llm-ollama/requirements.txt    # + streamlit, ollama

Sanity check:

python -c "import pandas, plotly, yaml, streamlit, ollama; print('OK')"

You should see OK.

3. Install Ollama

Windows (winget)

winget install Ollama.Ollama

Or grab the installer from ollama.com/download.

macOS / Linux

See ollama.com/download for the right package for your system.

Start the Ollama server

# In a separate terminal window — keep it open
ollama serve

Verify it’s responding:

curl http://127.0.0.1:11434/api/tags

You should see a JSON listing your models (empty at first run).

4. Pull the llama3.1:8b model

ollama pull llama3.1:8b

Downloads ~4.9 GB. Watch the progress. One-time only.

Verify:

ollama list

Expected:

NAME          ID              SIZE      MODIFIED
llama3.1:8b   46e0c10c039e    4.9 GB    a few minutes ago

Why llama3.1:8b?

We need a model that supports tool calling (calling query_sql cleanly) and fits in 8 GB of RAM. Good candidates: llama3.1:8b, qwen2.5:7b, mistral. Avoid llama3.2:3b, gemma3:270m — they don’t handle tools reliably.

5. Smoke test

From project-1/, run a one-shot pipeline test:

python csv-llm-shared/smoke_pipeline.py

This:

1. reads csv-llm-ollama/data/data1-anonymized.csv,
2. normalizes the rows,
3. inserts them into a local SQLite DB,
4. runs a SELECT to confirm everything is queryable.

Expected last line:

OK : 734 rows inserted, 1 query executed.

And to test Ollama on its own (no Streamlit):

python csv-llm-ollama/smoke_chat.py

Expected output (after 10–60 s):

Q : What is my biggest spending category?
Answered in 12.4s, 1 SQL call(s):
Your biggest spending category is Housing with a total of $30,895.29.
OK

If both pass, you’re ready for chapter 04.

Troubleshooting

Error	Likely cause	Fix
ConnectionError: Failed to connect to Ollama	ollama serve not running	Open a terminal and run ollama serve
model "llama3.1:8b" not found	Not pulled yet	ollama pull llama3.1:8b
ModuleNotFoundError: streamlit	Pip install missed	pip install -r csv-llm-ollama/requirements.txt
Permission denied on Activate.ps1	PowerShell execution policy	Set-ExecutionPolicy -Scope CurrentUser RemoteSigned
python not recognized (Windows)	Python not in PATH	Reinstall with “Add to PATH” checked

Takeaways

• Two git clone commands, two pip install, one ollama pull — that’s the whole setup.
• ollama serve must stay running in its own terminal.
• smoke_pipeline.py validates the data path; smoke_chat.py validates the LLM path.
• Total disk: ~5.5 GB (mostly the model).

Next: CSV ingestion across banks →