Add offline unit tests, CI workflow, MIT license, and metrics-honesty note

Author: jayaramdas07

.github/workflows/tests.yml

@@ -0,0 +1,28 @@
+name: tests
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+    branches: [main]
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+
+      - name: Install dependencies
+        run: |
+          python -m pip install --upgrade pip
+          pip install -r requirements.txt
+
+      - name: Run tests
+        env:
+          GROQ_API_KEY: test_dummy_key
+        run: python -m unittest discover -s tests -v

LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 K Jayarama Das
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

README.md

@@ -1,5 +1,8 @@
 # EcoPrompt — Energy-Efficient AI Prompt Routing
 
+![tests](https://github.com/jayaram-07/ecoprompt/actions/workflows/tests.yml/badge.svg)
+![license](https://img.shields.io/badge/license-MIT-green)
+
 > A hierarchical router that answers each prompt with the **cheapest, lowest-energy engine that can do the job** — sending trivial queries to deterministic/local handlers and reserving large LLMs only for prompts that genuinely need them. The result: lower latency, lower cost, and less compute/carbon per query.
 
 **🔗 [Live demo](https://frontend-two-indol-16.vercel.app/)**  ·  Frontend on Vercel · Backend on Google Cloud Run
@@ -49,6 +52,14 @@ Every request records latency, estimated energy (kWh), and estimated cost per ro
 - Groq Llama 3 70B: ~$0.70 / 1M tokens
 - Electricity: ₹8.00 / kWh (India avg)
 
+> **A note on the numbers (honesty matters).** The energy and CO₂ figures are
+> **estimates, not hardware measurements.** Energy is modeled as
+> `latency × assumed power draw` and cost is derived from the published
+> per-token prices above. They're meant to illustrate the *relative* savings of
+> routing cheap-first — not to be billed against. The one thing measured
+> directly is **cloud-avoidance rate** (the share of prompts answered without a
+> paid LLM call), which is the metric that actually drives the savings.
+
 ## Tech stack
 
 **Backend** — Python, FastAPI, Uvicorn · [Groq](https://groq.com) (Llama 3.1 8B / Llama 3 70B) · Google Gemini (grounded search) · custom deterministic + RAG engines
@@ -85,13 +96,26 @@ See [`.env.example`](.env.example). You'll need:
 - `GROQ_API_KEY` — from [console.groq.com](https://console.groq.com) (powers the local/large LLM tiers)
 - `GEMINI_API_KEY` — from [Google AI Studio](https://aistudio.google.com) (powers the grounded web-search tier)
 
+## Tests
+Offline unit tests cover the routing-decision logic (complexity scoring, the
+simple-prompt fast path, token budgeting, source-host matching, energy estimate)
+and the KB tokenizer. They make no API calls.
+
+```bash
+python -m unittest discover -s tests -v
+```
+
+CI runs them on every push via GitHub Actions (see the badge above).
+
 ## Project layout
 ```
 main.py              FastAPI app — routing cascade, metrics, streaming
 deterministic.py     Tier-1 rule/lookup engine
 kb/                  Knowledge-base lookups + RAG engine (geography, math, science, …)
+tests/               Offline unit tests for routing + tokenizer logic
 diagrams/            Architecture diagrams (.png/.svg/.mmd)
 frontend/            Vite + React + Tailwind UI and metrics dashboard
+.github/workflows/   CI (runs the test suite)
 ```
 
 ## Roadmap

tests/conftest.py

@@ -0,0 +1,7 @@
+"""Test setup: provide a dummy API key so `import main` succeeds without real
+credentials. The routing-helper functions under test are pure and never make
+network calls, so a placeholder key is sufficient.
+"""
+import os
+
+os.environ.setdefault("GROQ_API_KEY", "test_dummy_key")

tests/test_kb_utils.py

@@ -0,0 +1,35 @@
+"""Unit tests for the knowledge-base tokenizer used by the local KB routes.
+
+The tokenizer normalizes prompts (lowercasing, stopword removal, synonym
+folding) so cheap local lookups can match curated knowledge without an LLM.
+Run with:  python -m unittest discover -s tests
+"""
+import os
+import sys
+import unittest
+
+sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+
+from kb.kb_utils import tokenize  # noqa: E402
+
+
+class Tokenize(unittest.TestCase):
+    def test_lowercases_and_removes_stopwords(self):
+        self.assertEqual(tokenize("What is the capital of France"), ["capital", "france"])
+
+    def test_strips_punctuation(self):
+        self.assertEqual(tokenize("Photosynthesis?"), ["photosynthesis"])
+
+    def test_synonym_money_to_currency(self):
+        self.assertIn("currency", tokenize("what is the money"))
+        self.assertNotIn("money", tokenize("what is the money"))
+
+    def test_synonym_leader_to_president(self):
+        self.assertEqual(tokenize("who is the leader of India"), ["president", "india"])
+
+    def test_empty_string_returns_empty_list(self):
+        self.assertEqual(tokenize(""), [])
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/test_routing_helpers.py

@@ -0,0 +1,105 @@
+"""Unit tests for EcoPrompt's pure routing-decision helpers.
+
+These cover the logic that decides *which tier* answers a prompt and how many
+tokens to budget — the core of the energy-efficiency story. They run offline
+(no API calls). Run with:  python -m unittest discover -s tests
+"""
+import os
+import sys
+import unittest
+
+# Dummy key so `import main` (which constructs a Groq client) succeeds offline.
+os.environ.setdefault("GROQ_API_KEY", "test_dummy_key")
+# Allow `import main` when tests are run from the repo root.
+sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+
+import main  # noqa: E402
+
+
+class PromptComplexityScore(unittest.TestCase):
+    def test_trivial_prompt_scores_zero(self):
+        self.assertEqual(main.prompt_complexity_score("hi"), 0)
+
+    def test_short_factual_prompt_is_low(self):
+        self.assertLess(main.prompt_complexity_score("what is photosynthesis"), 4)
+
+    def test_long_multiclause_prompt_is_high(self):
+        prompt = (
+            "Compare and contrast monolithic versus microservices architecture, "
+            "because I need a detailed analysis of the tradeoffs"
+        )
+        self.assertGreaterEqual(main.prompt_complexity_score(prompt), 4)
+
+    def test_realtime_keyword_raises_score(self):
+        # "latest"/"today" should push a prompt toward the heavier routes.
+        with_realtime = main.prompt_complexity_score("what is the latest news today")
+        without = main.prompt_complexity_score("what is the news")
+        self.assertGreater(with_realtime, without)
+
+
+class IsSimplePrompt(unittest.TestCase):
+    def test_simple_definitional_prompt(self):
+        self.assertTrue(main.is_simple_prompt("what is photosynthesis"))
+
+    def test_complex_prompt_is_not_simple(self):
+        self.assertFalse(
+            main.is_simple_prompt(
+                "Compare and contrast microservices versus monolith in detail"
+            )
+        )
+
+    def test_long_prompt_is_not_simple(self):
+        long_prompt = "explain " + "word " * 20
+        self.assertFalse(main.is_simple_prompt(long_prompt))
+
+
+class SelectMaxTokens(unittest.TestCase):
+    def test_code_request_gets_largest_budget(self):
+        self.assertEqual(
+            main.select_max_tokens("write a python function to sort a list"), 420
+        )
+
+    def test_simple_prompt_gets_smallest_budget(self):
+        self.assertEqual(main.select_max_tokens("what is photosynthesis"), 120)
+
+    def test_complex_prompt_gets_large_budget(self):
+        prompt = (
+            "Compare and contrast monolithic versus microservices architecture, "
+            "because I need a detailed analysis of the tradeoffs"
+        )
+        self.assertEqual(main.select_max_tokens(prompt), 400)
+
+    def test_default_budget(self):
+        self.assertEqual(main.select_max_tokens("tell me a fun fact about dogs"), 280)
+
+
+class HostnameHelpers(unittest.TestCase):
+    def test_extract_hostname_strips_www_and_lowercases(self):
+        self.assertEqual(main.extract_hostname("https://www.BBC.com/news/x"), "bbc.com")
+
+    def test_extract_hostname_handles_bad_url(self):
+        self.assertEqual(main.extract_hostname("not a url"), "")
+
+    def test_hostname_matches_exact(self):
+        self.assertTrue(main.hostname_matches("bbc.com", "bbc.com"))
+
+    def test_hostname_matches_subdomain(self):
+        self.assertTrue(main.hostname_matches("news.bbc.com", "bbc.com"))
+
+    def test_hostname_does_not_match_unrelated(self):
+        self.assertFalse(main.hostname_matches("example.com", "bbc.com"))
+
+
+class EnergyEstimate(unittest.TestCase):
+    def test_energy_formula(self):
+        # ((ms/1000) * watts) / 3600 / 1000
+        self.assertAlmostEqual(
+            main.estimate_energy_kwh(1000, 50), (1 * 50) / 3600 / 1000, places=12
+        )
+
+    def test_zero_latency_is_zero_energy(self):
+        self.assertEqual(main.estimate_energy_kwh(0, 50), 0)
+
+
+if __name__ == "__main__":
+    unittest.main()