KK2 - Climate AI and Data Analysis

Typed LLM chain with FastAPI, Pandas, and SmolLM.

Requirements checklist

Requirement	Status
GET /health	Done
POST /data/upload (CSV, metadata)	Done
GET /data/stats (describe(), 404 if missing)	Done
POST /ai/ask (Runnable chain)	Done
Runnable chain PromptBuilder | LLMRunner | ResponseParser	Done
Pydantic types in each step	Done
SmolLM via transformers.pipeline	Done
.env not in git	Done
Error handling + meaningful status codes	Done
pytest (chain, endpoints, mocked model)	Done
reflektion.md	Done

Install

uv sync

Copy optional env file:

copy .env.example .env

Run

uv run uvicorn app.main:app --reload

Swagger: http://127.0.0.1:8000/docs

Example flow

1. Upload CSV:

curl -X POST "http://127.0.0.1:8000/data/upload" -F "file=@app/data/climate.csv"

2. Stats:

curl "http://127.0.0.1:8000/data/stats"

3. Ask:

curl -X POST "http://127.0.0.1:8000/ai/ask" \
  -H "Content-Type: application/json" \
  -d "{\"question\":\"Vilken stad har hogst medeltemperatur?\"}"

Without upload:

• /data/stats -> 404
• /ai/ask -> 400

Project structure

app/
  main.py
  config.py
  schemas.py
  data.py
  llm/
    runnable.py
    steps.py
    pipeline.py
  tests/
  data/
    climate.csv
reflektion.md

Tests

uv run pytest app/tests/ -v

Assumptions

• Dataset stored in memory (single process).
• CSV must be UTF-8.
• First /ai/ask call may be slow while model downloads (~300 MB).