wired up voice and vision, testing motion

r1/include/r1_voice/conversation.hh

@@ -5,6 +5,7 @@
 #include "r1_voice/llm.hh"
 #include "r1_voice/stt.hh"
 #include "r1_voice/tts.hh"
+#include "r1_voice/vision_context.hh"
 
 namespace r1_voice {
 
@@ -12,6 +13,7 @@ struct ConversationConfig {
     SttConfig stt;
     LlmConfig llm;
     TtsConfig tts;
+    VisionConfig vision;
 };
 
 class Conversation {
@@ -32,6 +34,7 @@ private:
     SpeechToText stt_;
     LLM llm_;
     TextToSpeech tts_;
+    VisionContext vision_;
 };
 
 } // namespace r1_voice

r1/include/r1_voice/llm.hh

@@ -33,7 +33,8 @@ public:
     explicit LLM(LlmConfig config);
 
     std::string chat(
-        const std::string& user
+        const std::string& user,
+        const std::string& transient_context = {}
     );
 
 private:

r1/include/r1_voice/vision_context.hh

@@ -0,0 +1,37 @@
+#pragma once
+
+#include <string>
+
+namespace r1_voice {
+
+struct VisionConfig {
+    std::string detections_path =
+        "/dev/shm/r1_detections.json";
+
+    // Reject YOLO output older than this.
+    double max_age_seconds = 2.5;
+
+    double min_confidence = 0.35;
+};
+
+class VisionContext {
+public:
+    explicit VisionContext(
+        VisionConfig config
+    );
+
+    bool isVisionQuestion(
+        const std::string& text
+    ) const;
+
+    std::string buildContext() const;
+
+private:
+    VisionConfig config_;
+
+    static std::string lowerAscii(
+        std::string text
+    );
+};
+
+} // namespace r1_voice