truth_serum_academic 0.1.0 → 0.2.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 61567f2d575ced3d37904bc6c64d1c4087b44a545082d13a7ba4dff40bb8a30c
-  data.tar.gz: c2a4782c0349682575993feaea7de717409ce8cbbea7baf348b629d9e9d8a097
+  metadata.gz: 0aa6c1317c474b2de311edc851b77ac74fbeaee5020106506fee70345be91b42
+  data.tar.gz: 35ed231086b07be3a4b1ce972291a7a8883995e620c6c07d07e3188c9aa71449
 SHA512:
-  metadata.gz: 00ebb740e006b66f5b83c25fea94333e06a2dcb155bb46736bbebd18fbda4df786c3792acc36709812240d5df7cff48e3ad7e721f5795c4cc83c6a6866a95076
-  data.tar.gz: c25c9d9db9092e94db7725e82ec6379196814b6e3fceaf8dccc79756b1c51be51ab9a84223280fa0f21864a5272d419621b21dc79ff74f97026c34ab5cdf70f7
+  metadata.gz: 14d7028af3a45ae091512aa7c3259d3908e79d2cacd37ae577935ca03d0e86cd15d2a1fd8eab483012ca7a6e8238ea8349a453904abb7f1b3a3786c0014eaeab
+  data.tar.gz: b977543f46a062c0397b0976db541115a3f69314925f5af1f0cadfc2fce01cb9bdf33cf648b3e075d9ec176d9a6495d12315e9b58c612928459af8f8d7504ce3

data/lib/truth_serum/verifier.rb

@@ -1,56 +1,78 @@
 require 'httparty'
+require 'json'
 require_relative 'result'
 
 module TruthSerum
   class Verifier
-    BASE_URL = "https://api.crossref.org/works"
+    # We use the fast, free-tier friendly model
+    API_URL = "https://generativelanguage.googleapis.com/v1beta/models/gemini-1.5-flash:generateContent"
 
     def self.check_list(list)
+      # 1. SECURITY CHECK
+      # We look for the key in your computer's environment variables.
+      # NEVER hardcode your key here if you publish the gem!
+      api_key = ENV["GEMINI_API_KEY"]
+      
+      if api_key.nil? || api_key.empty?
+        puts "⚠️  ERROR: Missing API Key."
+        puts "Please run: set GEMINI_API_KEY=your_key_here"
+        exit
+      end
+
       results = []
 
       list.each do |citation|
-        response = HTTParty.get(BASE_URL, query: { 
-          "query.bibliographic" => citation, 
-          "rows" => 1 
-        })
-
-        data = response.parsed_response
-        items = data.dig("message", "items")
-
-        status = :fake
-        score = 0.0
-        
-        # LOGIC: Check if we found a match AND if it looks similar
-        if items && items.any?
-          top_match = items.first
-          found_title = top_match["title"]&.first || ""
+        # 2. CONSTRUCT THE PROMPT
+        # We give Gemini strict instructions to act like a JSON machine.
+        prompt_text = <<~TEXT
+          You are a strict academic verification tool. Check if the following citation is real or a hallucination.
+          Citation: "#{citation}"
           
-          # 1. Normalize strings (lowercase, find words 4+ letters long)
-          # We ignore small words like "the", "of", "and".
-          input_words = citation.downcase.scan(/[a-z]{4,}/)
-          found_words = found_title.downcase.scan(/[a-z]{4,}/)
+          Respond ONLY with valid JSON in this format:
+          {
+            "is_real": boolean,
+            "confidence_score": number_between_0_and_1,
+            "reasoning": "short explanation"
+          }
+        TEXT
+
+        # 3. SEND TO GEMINI
+        response = HTTParty.post(
+          "#{API_URL}?key=#{api_key}",
+          headers: { 'Content-Type' => 'application/json' },
+          body: {
+            contents: [{ parts: [{ text: prompt_text }] }]
+          }.to_json
+        )
+
+        # 4. PARSE THE BRAIN WAVES (JSON)
+        begin
+          payload = response.parsed_response
+          # Dig through Gemini's nested JSON response structure
+          raw_text = payload.dig("candidates", 0, "content", "parts", 0, "text")
           
-          # 2. Count common words
-          common_words = input_words & found_words
+          # Clean up any markdown formatting (Gemini likes to wrap JSON in ```json ... ```)
+          clean_json = raw_text.gsub(/```json/, "").gsub(/```/, "")
+          data = JSON.parse(clean_json)
+
+          # 5. CREATE RESULT
+          status = data["is_real"] ? :real : :fake
+          score = data["confidence_score"]
           
-          # 3. Decision Threshold
-          # If they share at least 2 significant words, we assume it's the same paper.
-          if common_words.size >= 2
-            status = :real
-            score = top_match["score"]
-          else
-            # We found a paper, but the title was too different.
-            # (e.g. You searched "Musk Mars" and got "Geology of Mars")
-            status = :fake 
-            score = 0.1 # Low score for mismatch
-          end
-        end
+          # We can even print the reasoning if we want!
+          # For now, we stick to the score/status structure.
+          result = TruthSerum::Result.new(
+            text: citation,
+            score: score,
+            status: status
+          )
+          results << result
 
-        results << TruthSerum::Result.new(
-          text: citation,
-          score: score,
-          status: status
-        )
+        rescue => e
+          # If Gemini gets confused or the internet breaks
+          puts "Error parsing Gemini response: #{e.message}"
+          results << TruthSerum::Result.new(text: citation, score: 0.0, status: :fake)
+        end
       end
 
       results

data/lib/truth_serum/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module TruthSerum
-  VERSION = "0.1.0"
+  VERSION = "0.2.0"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: truth_serum_academic
 version: !ruby/object:Gem::Version
-  version: 0.1.0
+  version: 0.2.0
 platform: ruby
 authors:
 - Nathalie Tasler