Varijabla

Overview

QED is a Wolfram Alpha-style computational knowledge engine: natural language in, structured deterministic answers out.

Core philosophy

  • Zero hallucination — LLMs never compute math or generate factual answers.
  • Deterministic execution — all math, physics and fact retrieval go through dedicated tools and APIs.
  • LLM as planner — the LLM strictly orchestrates backend tools for complex queries; it does not compute.
  • Assumptions labelled — Fermi-style estimates are always flagged, never silently presented as fact.

Services

QED runs as three services via Docker Compose:

Service Stack Port Description
qed-orchestrator Flask / gunicorn 8850 → 5000 NLP pipeline, routing, caching, response assembly
qed-data FastAPI 8800 → 8000 Compute backends: math, chemistry, weather, databases, Wikidata
qed-redis redis:7-alpine — Query cache and rate-limiter storage

The orchestrator understands queries in five languages (English, German, French, Spanish, Croatian) and routes them through a three-stage NLP waterfall — see Architecture.

Supported domains

# Domain Backend
1 Basic math & unit conversions qalc subprocess
2 Symbolic math (Wolfram Language notation) qalc via WL→qalc translator
3 Plotting sympy + matplotlib
4 Chemistry — balancing & elements chempy + mendeleev
5 Weather Open-Meteo API + geocoding
6 Geography / facts Wikidata API + Wikipedia
7 Nutrition SQLite food database
8 Linguistics / dictionary Wikipedia + LLM fallback; Croatian via HJP
9 Temporal logic python-dateutil

Example queries

Query Result
30% of 8 miles 2.4 miles
100 m/s to mph 223.7 mph
Integrate[x^2, x] LaTeX: x³/3
balance H2 + O2 -> H2O 2H₂ + O₂ → 2H₂O
molar mass of carbon 12.011 g/mol
weather in Zagreb tomorrow live Open-Meteo data
what day was Christmas 1992 Friday
Plot[Sin[x], {x, 0, 2Pi}] rendered PNG
capital of Croatia Zagreb (Wikidata)
calories in a lightyear of fried chicken 2.365×10¹⁹ kcal (flagged Fermi estimate)

Where to go next