Мен бұл $40 ESP32 тақтасына схема сыздым және AI оны түсіндірді

/*
 * ===========================================================================
 *  08_Draw_AI  —  MaTouch AI ESP32-S3 2.8" TFT ST7789V
 * ===========================================================================
 *
 *  ---------------------------------------------------------------------------
 *  ROBOJAX.COM  -  MaTouch AI ESP32-S3 2.8" project series
 *
 *    WATCH THE Introduction and face recognition VIDEO
 *        https://youtu.be/6AL3g3tC_Hk
 *    Video 2: Draw AI - https://youtu.be/FaIW1MVIsj4
 *
 *    WRITTEN TUTORIALS - every project, with photos and full explanation
 *        Camera and touchscreen.... https://robojax.com/RTJ849
 *        Offline face recognition.. https://robojax.com/RTJ850
 *        AI voice assistant........ https://robojax.com/RTJ851
 *        AI vision................. https://robojax.com/RTJ852
 *        Draw AI  ................. https://robojax.com/RTJ867 (this code)
 *
 *    GET THE BOARD - SAVE $5 with coupon code:  Robojax_Makerfab
 *        https://www.makerfabs.com/matouch-ai-esp32s3-2-8-tft-st7789v.html
 *        (enter the code at checkout)
 *
 *  All of this code is free. If it helped you, a subscribe on YouTube is
 *  the best way to support more of it.
 *  ---------------------------------------------------------------------------
 *
 *  Draw or write on the touchscreen with your finger, tap ASK, and an AI
 *  tells you what you drew - on the screen and out loud.
 *
 *  THREE MODES (tap MODE to cycle):
 *
 *      WHAT IS THIS?    draw a house, a cat, a triangle -> it names it
 *      READ THIS        write HELLO in big letters -> it reads it back
 *      EXPLAIN CIRCUIT  sketch a battery, resistor and LED -> it tells you
 *                       what the components are and what the circuit does
 *
 *  That last mode is the interesting one: a hand-drawn schematic, explained
 *  by an AI, on a forty-dollar board with no computer attached.
 *
 *  NO CAMERA IS USED by this sketch. That means the J1/J3 Mabee sockets are
 *  free GPIO here (see 07_GPIO_Test), and there is no camera/SD memory
 *  contention to worry about.
 *
 *  ---------------------------------------------------------------------------
 *  HOW IT WORKS (the trick worth knowing)
 *
 *  You cannot read pixels back out of the ST7789 over SPI, and Arduino_GFX
 *  normally draws straight to the panel with no copy in memory - so there is
 *  no "screenshot" to send anywhere. The fix: draw into an Arduino_Canvas,
 *  which keeps a full framebuffer in PSRAM and flushes it to the display.
 *  That buffer IS the image, so fmt2jpg() turns it straight into a JPEG and
 *  the rest of the path is identical to 05_Vision_AI.
 *
 *  Black ink on a white canvas, deliberately: vision models read dark-on-light
 *  far more reliably than the reverse, and it looks like paper on camera.
 *
 *  ---------------------------------------------------------------------------
 *  TIPS
 *    - A cheap capacitive stylus is MUCH better than a fingertip.
 *    - Write BIG. The canvas is 240x240 - a few large characters, not a
 *      sentence.
 *    - There is no palm rejection: only the first touch point is used, so
 *      keep your palm off the glass.
 *
 *  ---------------------------------------------------------------------------
 *  FILL IN secrets.h BEFORE FLASHING (WIFI_* and OPENAI_*; AZURE_* only if
 *  SPEAK_REPLIES is 1).
 *
 *  BOARD SETTINGS (Tools menu - EVERY line matters)
 *      Board            : ESP32S3 Dev Module
 *      ESP32 core       : 2.0.17
 *      PSRAM            : OPI PSRAM        <-- required, the canvas lives there
 *      Flash Size       : 16MB (128Mb)
 *      Partition Scheme : 16M Flash (3MB APP/9.9MB FATFS)
 *      USB CDC On Boot  : Disabled
 *      Upload Speed     : 921600
 *      Port             : the CH340K USB-C port (the one near RESET)
 *
 *  LIBRARIES
 *      GFX Library for Arduino   v1.5.6   (NOT 1.6.x - that pairs with core 3)
 *      bb_captouch               v1.3.1
 *      ArduinoJson               v7.x
 *      Adafruit NeoPixel         any recent
 *
 *  ---------------------------------------------------------------------------
 *  FUNCTIONS IN THIS SKETCH
 *      led(r,g,b)          set the RGB status LED colour
 *      getTouch(&x,&y)     read the touch panel, mapped to screen coordinates
 *      clearCanvas()       wipe to white and redraw the hint
 *      drawStroke(...)     one thick line segment, into the canvas AND the panel
 *      handleDrawing(...)  turn touch samples into continuous strokes
 *      waitData()/readHttpResponse()   robust HTTPS reply reader (de-chunking)
 *      canvasToJpeg(&len)  RGB565 canvas -> JPEG in PSRAM
 *      askVision(...)      send the drawing + prompt to OpenAI, get an answer
 *      spkInit()/speak()/playLastAnswer()   Azure voice out, with free REPLAY
 *      drawButtons()/drawAnswerSide()/showAnswer()/drawWifi()   the UI
 *      askAboutDrawing()   one full cycle: encode -> ask -> show -> speak
 *      setup() / loop()    boot + WiFi / drawing and button handling
 *
 *  Robojax.com
 * ===========================================================================
 */

#define SPEAK_REPLIES  1     // 1 = read the answer aloud with Azure TTS
#define LED_BRIGHTNESS 15    // WS2812 is on the power rail - 15 is plenty

#include <Arduino_GFX_Library.h>
#include <bb_captouch.h>
#include <Adafruit_NeoPixel.h>
#include <ArduinoJson.h>
#include <WiFi.h>
#include <WiFiClientSecure.h>
#include "mbedtls/base64.h"
#include "img_converters.h"      // fmt2jpg()
#include "esp_camera.h"          // for the pixformat_t enum used by fmt2jpg
#include "pins.h"
#include "secrets.h"

#if SPEAK_REPLIES
#include "driver/i2s.h"
#define WAV_HEADER_LEN 44
#endif

Arduino_HWSPI *bus = new Arduino_HWSPI(
    TFT_DC, TFT_CS, TFT_SCLK, TFT_MOSI, TFT_MISO, &SPI, true);
Arduino_GFX *panel = new Arduino_ST7789(bus, TFT_RES, 1, true);

/* The canvas is what we actually draw into: a real framebuffer in PSRAM that
 * we can both flush to the panel AND hand to the JPEG encoder. */
#define CANVAS_W 240
#define CANVAS_H 240
Arduino_Canvas *canvas = new Arduino_Canvas(CANVAS_W, CANVAS_H, panel, 0, 0);

BBCapTouch bbct;
Adafruit_NeoPixel rgb(RGB_LED_NUM, RGB_LED_PIN, NEO_GRB + NEO_KHZ800);

/* --- the three modes ------------------------------------------------------ */
struct DrawMode { const char *l1, *l2; const char *prompt; };
const DrawMode MODES[] = {
  {"WHAT IS", "THIS?",
   "This is a drawing made with a finger on a small touchscreen. Name the "
   "shape or object in ONE short sentence of at most 15 words. Plain text only."},
  {"READ",    "THIS",
   "This image contains handwriting drawn with a finger. Reply with ONLY the "
   "text you can read, nothing else. If you cannot read it, say: cannot read it."},
  {"EXPLAIN", "CIRCUIT",
   "This is a hand-drawn electronic circuit diagram. In at most two short "
   "sentences, name the components you can identify and say what the circuit "
   "does. Plain text only. If it is not a circuit, say so."},
};
#define N_MODES 3
int mode = 0;

/* --- layout: 240 canvas on the left, buttons on the right ----------------- */
#define BTN_X   244
#define BTN_W    72
#define BTN_H    46
#define BTN_MODE_Y   4
#define BTN_ASK_Y   56
#define BTN_CLEAR_Y 108

bool showing_answer = false;

/* --- drawing state -------------------------------------------------------- */
int  last_x = -1, last_y = -1;
bool has_ink = false;

/* --- kept audio for REPLAY ------------------------------------------------ */
uint8_t *last_audio = nullptr;
size_t   last_audio_len = 0;

void led(uint8_t r, uint8_t g, uint8_t b) { rgb.setPixelColor(0, rgb.Color(r, g, b)); rgb.show(); }
void playLastAnswer();
void drawButtons();


/* ===========================================================================
 *  Touch
 * =========================================================================== */
bool getTouch(uint16_t *x, uint16_t *y) {
  TOUCHINFO ti;
  if (!bbct.getSamples(&ti)) return false;
  if (ti.count < 1) return false;            // first point only - no palm rejection
  *x = ti.y[0];
  *y = (ti.x[0] > 240) ? 0 : (240 - ti.x[0]);
  return true;
}


/* ===========================================================================
 *  Canvas / drawing
 * =========================================================================== */
void clearCanvas() {
  canvas->fillScreen(WHITE);
  canvas->flush();                                 // one full push: paint it white
  has_ink = false;
  last_x = last_y = -1;

  /* The hint goes on the PANEL, never into the canvas - anything drawn into
   * the canvas becomes part of the image the AI sees, and it would happily
   * read this text back to you. */
  panel->setTextSize(1);
  panel->setTextColor(panel->color565(170, 170, 170));
  panel->setCursor(6, CANVAS_H - 12);
  panel->print("draw here, then tap ASK");
}

/* One thick line segment, drawn into BOTH the canvas and the panel.
 *
 * Why both: the canvas is the image we will send to the AI, but flushing the
 * whole 240x240 buffer to the screen costs ~115 KB of SPI. Doing that per
 * touch sample throttled the loop so hard that most samples were missed -
 * which is exactly what "it only draws dots" looks like. Painting the same
 * few pixels directly to the panel is essentially free, so the canvas is only
 * flushed when the screen genuinely has to be rebuilt. */
static const int8_t STROKE_OFFS[5][2] = {{0,0},{1,0},{-1,0},{0,1},{0,-1}};

void drawStroke(int x0, int y0, int x1, int y1) {
  for (int i = 0; i < 5; i++) {
    int ox = STROKE_OFFS[i][0], oy = STROKE_OFFS[i][1];
    canvas->drawLine(x0 + ox, y0 + oy, x1 + ox, y1 + oy, BLACK);
    panel->drawLine(x0 + ox, y0 + oy, x1 + ox, y1 + oy, BLACK);
  }
}

/* Touch is read once per loop and handed in here.
 *
 * THE IMPORTANT BIT - why this is time-based rather than "no touch = lifted":
 * the GT911 does not return a point on every poll. It reports when its own
 * internal sample is ready, so a poll in between comes back empty even though
 * your finger never left the glass. Treating each empty poll as a lift resets
 * the stroke, and every real sample then lands as an isolated dot instead of
 * a line joined to the previous one - which is exactly what "it only draws
 * dots" looks like. So a stroke only ends after the panel has been quiet for
 * LIFT_MS, which bridges the dropouts but is still far shorter than a genuine
 * lift-and-reposition. */
#define LIFT_MS 70

void handleDrawing(bool touched, int x, int y) {
  static uint32_t last_seen = 0;

  if (touched && x >= 0 && x < CANVAS_W) {
    if (last_x >= 0) drawStroke(last_x, last_y, x, y);   // join to the last point
    else             drawStroke(x, y, x, y);             // start of a stroke
    last_x = x; last_y = y;
    last_seen = millis();
    has_ink = true;
  } else if (last_x >= 0 && millis() - last_seen > LIFT_MS) {
    last_x = last_y = -1;                                // really lifted now
  }
}


/* ===========================================================================
 *  HTTP  —  the same robust reader proven in 04 and 05
 * =========================================================================== */
static bool waitData(WiFiClientSecure &c, uint32_t ms) {
  uint32_t t0 = millis();
  while (!c.available()) {
    if (!c.connected()) return false;
    if (millis() - t0 > ms) return false;
    delay(10);
  }
  return true;
}

static int readHttpResponse(WiFiClientSecure &client, String &body_out, uint32_t idle_ms) {
  body_out = "";
  if (!waitData(client, idle_ms)) { Serial.println("HTTP: no response at all"); return 0; }

  String status_line = client.readStringUntil('\n');
  int code = 0;
  sscanf(status_line.c_str(), "HTTP/%*s %d", &code);

  bool chunked = false;
  while (waitData(client, idle_ms)) {
    String h = client.readStringUntil('\n');
    if (h == "\r" || h.length() <= 1) break;
    h.toLowerCase();
    if (h.startsWith("transfer-encoding:") && h.indexOf("chunked") >= 0) chunked = true;
  }

  if (chunked) {
    int blanks = 0;
    while (true) {
      if (!waitData(client, idle_ms)) { Serial.println("HTTP: chunk timeout"); break; }
      String szline = client.readStringUntil('\n');
      szline.trim();
      if (szline.length() == 0) { if (++blanks > 4) break; continue; }
      blanks = 0;
      long sz = strtol(szline.c_str(), NULL, 16);
      if (sz <= 0) break;
      long got = 0;
      while (got < sz) {
        if (!waitData(client, idle_ms)) break;
        while (client.available() && got < sz) { body_out += (char)client.read(); got++; }
      }
      if (waitData(client, 3000)) client.readStringUntil('\n');
      if (got < sz) break;
    }
  } else {
    while (waitData(client, idle_ms))
      while (client.available()) body_out += (char)client.read();
  }
  return code;
}


/* ===========================================================================
 *  Canvas -> JPEG
 *  The canvas framebuffer is plain RGB565, which fmt2jpg() accepts directly.
 * =========================================================================== */
uint8_t *canvasToJpeg(size_t *len_out) {
  *len_out = 0;
  uint16_t *fb = canvas->getFramebuffer();
  if (!fb) { Serial.println("no canvas framebuffer"); return nullptr; }

  uint8_t *jpg = NULL;
  size_t jpg_len = 0;
  if (!fmt2jpg((uint8_t *)fb, (size_t)CANVAS_W * CANVAS_H * 2,
               CANVAS_W, CANVAS_H, PIXFORMAT_RGB565, 90, &jpg, &jpg_len)) {
    Serial.println("fmt2jpg failed");
    return nullptr;
  }
  *len_out = jpg_len;
  return jpg;                                  // caller frees
}


/* ===========================================================================
 *  OpenAI vision  —  identical shape to 05_Vision_AI
 * =========================================================================== */
bool askVision(const uint8_t *jpg, size_t jpg_len, const char *prompt, String &answer_out) {
  size_t b64_cap = ((jpg_len + 2) / 3) * 4 + 16;
  unsigned char *b64 = (unsigned char *)ps_malloc(b64_cap);
  if (!b64) return false;
  size_t b64_len = 0;
  if (mbedtls_base64_encode(b64, b64_cap, &b64_len, jpg, jpg_len) != 0) { free(b64); return false; }

  const char *head_fmt =
      "{\"model\":\"%s\",\"max_completion_tokens\":%d,\"messages\":[{\"role\":\"user\","
      "\"content\":[{\"type\":\"text\",\"text\":\"%s\"},"
      "{\"type\":\"image_url\",\"image_url\":{\"url\":\"data:image/jpeg;base64,";
  const char *tail = "\"}}]}]}";

  char head[900];
  int head_len = snprintf(head, sizeof(head), head_fmt, OPENAI_MODEL, LLM_MAX_TOKENS, prompt);

  size_t body_len = head_len + b64_len + strlen(tail);
  char *body = (char *)ps_malloc(body_len + 1);
  if (!body) { free(b64); return false; }
  memcpy(body, head, head_len);
  memcpy(body + head_len, b64, b64_len);
  strcpy(body + head_len + b64_len, tail);
  free(b64);

  WiFiClientSecure client;
  client.setInsecure();
  client.setTimeout(45);

  if (!client.connect(OPENAI_HOST, 443)) {
    Serial.println("Vision: TLS connect failed");
    free(body);
    return false;
  }

  client.print(String("POST /v1/chat/completions HTTP/1.1\r\n"
               "Host: " OPENAI_HOST "\r\n"
               "Authorization: Bearer " OPENAI_KEY "\r\n"
               "Content-Type: application/json\r\n"
               "Connection: close\r\n"
               "Content-Length: ") + String(body_len) + "\r\n\r\n");

  size_t sent = 0;
  while (sent < body_len) {
    size_t n = min((size_t)4096, body_len - sent);
    size_t w = client.write((uint8_t *)body + sent, n);
    if (w == 0) { delay(50); w = client.write((uint8_t *)body + sent, n); if (w == 0) break; }
    sent += w;
    yield();
  }
  free(body);
  if (sent < body_len) { client.stop(); Serial.println("Vision: upload stalled"); return false; }

  String resp;
  int code = readHttpResponse(client, resp, 45000);
  client.stop();

  if (code != 200) { Serial.printf("Vision HTTP %d: %s\n", code, resp.c_str()); return false; }

  JsonDocument doc;
  if (deserializeJson(doc, resp)) {
    Serial.printf("Vision: JSON parse failed, %u bytes\n", resp.length());
    return false;
  }
  const char *content = doc["choices"][0]["message"]["content"];
  if (!content) return false;
  answer_out = String(content);
  answer_out.trim();
  return answer_out.length() > 0;
}


/* ===========================================================================
 *  Azure voice out  —  same de-chunked player as 04/05
 * =========================================================================== */
#if SPEAK_REPLIES
void spkInit() {
  i2s_config_t cfg = {
    .mode = (i2s_mode_t)(I2S_MODE_MASTER | I2S_MODE_TX),
    .sample_rate = 16000,
    .bits_per_sample = I2S_BITS_PER_SAMPLE_16BIT,
    .channel_format = I2S_CHANNEL_FMT_ONLY_LEFT,
    .communication_format = I2S_COMM_FORMAT_STAND_I2S,
    .intr_alloc_flags = ESP_INTR_FLAG_LEVEL1,
    .dma_buf_count = 8,
    .dma_buf_len = 256,
    .use_apll = false,
    .tx_desc_auto_clear = true,
    .fixed_mclk = 0
  };
  i2s_pin_config_t pins = {
    .mck_io_num = I2S_PIN_NO_CHANGE,
    .bck_io_num = I2S_SPK_BCLK,
    .ws_io_num = I2S_SPK_LRC,
    .data_out_num = I2S_SPK_DOUT,
    .data_in_num = I2S_PIN_NO_CHANGE
  };
  i2s_driver_install(I2S_SPK_PORT, &cfg, 0, NULL);
  i2s_set_pin(I2S_SPK_PORT, &pins);
  i2s_zero_dma_buffer(I2S_SPK_PORT);
}

void playLastAnswer() {
  if (!last_audio || last_audio_len <= WAV_HEADER_LEN) return;
  static const uint8_t lead_in[640] = {0};
  size_t w = 0;
  i2s_write(I2S_SPK_PORT, lead_in, sizeof(lead_in), &w, portMAX_DELAY);
  i2s_write(I2S_SPK_PORT, last_audio + WAV_HEADER_LEN,
            last_audio_len - WAV_HEADER_LEN, &w, portMAX_DELAY);
  i2s_write(I2S_SPK_PORT, lead_in, sizeof(lead_in), &w, portMAX_DELAY);
  delay(150);
  i2s_zero_dma_buffer(I2S_SPK_PORT);
}

/* Read exactly n bytes into a raw buffer. BINARY SAFE - unlike the String
 * based reader used for JSON, which cannot carry audio: PCM is full of 0x00
 * bytes, and an Arduino String treats those as end-of-string. */
static size_t readExact(WiFiClientSecure &c, uint8_t *dst, size_t n) {
  size_t got = 0;
  uint32_t t0 = millis();
  while (got < n && millis() - t0 < 10000) {
    int r = c.read(dst + got, n - got);
    if (r > 0) { got += r; t0 = millis(); }
    else if (!c.connected() && !c.available()) break;
    else delay(2);
  }
  return got;
}

void speak(const String &text) {
  String safe = text;
  safe.replace("&", "&amp;"); safe.replace("<", "&lt;"); safe.replace(">", "&gt;");
  String ssml = "<speak version='1.0' xml:lang='" AZURE_TTS_LANG "'>"
                "<voice name='" AZURE_TTS_VOICE "'>" + safe + "</voice></speak>";

  WiFiClientSecure client;
  client.setInsecure();
  client.setTimeout(30);

  const char *host = AZURE_REGION ".tts.speech.microsoft.com";
  if (!client.connect(host, 443)) { Serial.println("TTS: TLS connect failed"); return; }

  client.print(String("POST /cognitiveservices/v1 HTTP/1.1\r\nHost: ") + host + "\r\n"
               "Ocp-Apim-Subscription-Key: " AZURE_SPEECH_KEY "\r\n"
               "Content-Type: application/ssml+xml\r\n"
               "X-Microsoft-OutputFormat: riff-16khz-16bit-mono-pcm\r\n"
               "User-Agent: MaTouchRobojax\r\nConnection: close\r\n"
               "Content-Length: " + String(ssml.length()) + "\r\n\r\n");
  client.print(ssml);

  /* --- status line + headers (text, so String is fine here) --- */
  if (!waitData(client, 30000)) { Serial.println("TTS: no response"); client.stop(); return; }
  String status_line = client.readStringUntil('\n');
  int code = 0;
  sscanf(status_line.c_str(), "HTTP/%*s %d", &code);

  bool chunked = false;
  long content_len = -1;
  while (waitData(client, 30000)) {
    String h = client.readStringUntil('\n');
    if (h == "\r" || h.length() <= 1) break;
    h.toLowerCase();
    if (h.startsWith("transfer-encoding:") && h.indexOf("chunked") >= 0) chunked = true;
    if (h.startsWith("content-length:")) content_len = h.substring(15).toInt();
  }
  if (code != 200) { Serial.printf("TTS HTTP %d\n", code); client.stop(); return; }

  /* --- body straight into PSRAM as raw bytes --- */
  const size_t AUDIO_CAP = 1200 * 1024;          // ~37 s of 16 kHz mono
  uint8_t *audio = (uint8_t *)ps_malloc(AUDIO_CAP);
  if (!audio) { Serial.println("TTS: PSRAM alloc failed"); client.stop(); return; }
  size_t alen = 0;

  if (chunked) {
    int blanks = 0;
    while (alen < AUDIO_CAP) {
      if (!waitData(client, 15000)) break;
      String szline = client.readStringUntil('\n');
      szline.trim();
      if (szline.length() == 0) { if (++blanks > 4) break; continue; }
      blanks = 0;
      long sz = strtol(szline.c_str(), NULL, 16);
      if (sz <= 0) break;                                   // final chunk
      if ((size_t)sz > AUDIO_CAP - alen) sz = (long)(AUDIO_CAP - alen);
      size_t got = readExact(client, audio + alen, (size_t)sz);
      alen += got;
      if (waitData(client, 3000)) client.readStringUntil('\n');   // CRLF
      if (got < (size_t)sz) break;
    }
  } else if (content_len > 0) {
    size_t want = (size_t)content_len < AUDIO_CAP ? (size_t)content_len : AUDIO_CAP;
    alen = readExact(client, audio, want);
  } else {
    uint32_t idle = millis();
    while (millis() - idle < 5000 && alen < AUDIO_CAP) {
      int r = client.read(audio + alen, AUDIO_CAP - alen);
      if (r > 0) { alen += r; idle = millis(); }
      else if (!client.connected()) break;
      else delay(5);
    }
  }
  client.stop();

  if (alen <= WAV_HEADER_LEN) { Serial.println("TTS: no audio"); free(audio); return; }

  if (last_audio) free(last_audio);
  last_audio = audio;                            // kept for REPLAY
  last_audio_len = alen;

  Serial.printf("TTS: %u KB clean audio (%s), playing\n",
                (unsigned)(alen / 1024), chunked ? "de-chunked" : "plain");
  playLastAnswer();
}
#else
void spkInit() {}
void speak(const String &t) { (void)t; }
void playLastAnswer() {}
#endif


/* ===========================================================================
 *  UI  —  right-hand button column (drawn on the PANEL, not the canvas, so
 *  it is never part of the image sent to the AI)
 * =========================================================================== */
void drawWifi() {
  panel->fillRect(BTN_X, 196, BTN_W, 44, BLACK);
  bool up = (WiFi.status() == WL_CONNECTED);
  long rssi = up ? WiFi.RSSI() : -100;
  int bars = rssi > -55 ? 4 : rssi > -65 ? 3 : rssi > -75 ? 2 : rssi > -85 ? 1 : 0;
  for (int b = 0; b < 4; b++) {
    int bh = 5 + b * 5;
    panel->fillRect(BTN_X + 4 + b * 8, 224 - bh, 6, bh,
                    (b < bars) ? GREEN : panel->color565(60, 60, 60));
  }
  panel->setTextSize(1);
  panel->setCursor(BTN_X + 40, 210);
  panel->setTextColor(up ? CYAN : RED);
  if (up) panel->printf("%ld", rssi); else panel->print("--");
  panel->setCursor(BTN_X + 2, 230);
  panel->setTextColor(panel->color565(120, 120, 120));
  panel->print("Robojax.com");
}

void drawButton(int y, const char *l1, const char *l2, uint16_t colour) {
  panel->fillRoundRect(BTN_X, y, BTN_W, BTN_H, 6, colour);
  panel->drawRoundRect(BTN_X, y, BTN_W, BTN_H, 6, WHITE);
  panel->setTextSize(1);
  panel->setTextColor(WHITE);
  panel->setCursor(BTN_X + 6, y + 14);
  panel->print(l1);
  panel->setCursor(BTN_X + 6, y + 26);
  panel->print(l2);
}

void drawButtons() {
  panel->fillRect(BTN_X, 0, BTN_W, 196, BLACK);
  drawButton(BTN_MODE_Y,  MODES[mode].l1, MODES[mode].l2, panel->color565(90, 60, 140));
  drawButton(BTN_ASK_Y,   "ASK",  "THE AI",  panel->color565(0, 100, 60));
  drawButton(BTN_CLEAR_Y, "CLEAR", "",       panel->color565(110, 35, 35));
  panel->setTextColor(YELLOW);
  panel->setCursor(BTN_X + 2, 166);
  panel->print("MODE cycles");
  panel->setCursor(BTN_X + 2, 178);
  panel->print("the question");
  drawWifi();
  showing_answer = false;
}

void drawAnswerSide() {
  panel->fillRect(BTN_X, 0, BTN_W, 196, BLACK);
  drawButton(BTN_MODE_Y, "BACK", "TO DRAW", panel->color565(0, 90, 160));
#if SPEAK_REPLIES
  if (last_audio_len) drawButton(BTN_ASK_Y, "REPLAY", "", panel->color565(0, 110, 60));
#endif
  drawButton(BTN_CLEAR_Y, "CLEAR", "ALL", panel->color565(110, 35, 35));
  drawWifi();
  showing_answer = true;
}

/* The answer is drawn on the PANEL over the canvas area, and stays until the
 * user taps BACK or CLEAR - the canvas underneath is untouched. */
void showAnswer(const String &text) {
  const int max_chars = 38;
  int lines = (text.length() + max_chars - 1) / max_chars;
  if (lines > 8) lines = 8;
  int h = lines * 10 + 12;
  int y = CANVAS_H - h;

  panel->fillRect(0, y, CANVAS_W, h, BLACK);
  panel->drawRect(0, y, CANVAS_W, h, CYAN);
  panel->setTextSize(1);
  panel->setTextColor(WHITE);
  for (int i = 0; i < lines; i++) {
    panel->setCursor(4, y + 6 + i * 10);
    panel->print(text.substring(i * max_chars,
                                min((int)text.length(), (i + 1) * max_chars)));
  }
}


/* ===========================================================================
 *  One full ask cycle
 * =========================================================================== */
void askAboutDrawing() {
  if (!has_ink) {
    showAnswer("Draw something first, then tap ASK.");
    drawAnswerSide();
    return;
  }

  led(255, 120, 0);                                   // amber: working
  panel->fillRect(0, CANVAS_H - 22, CANVAS_W, 22, BLACK);
  panel->setTextSize(1);
  panel->setTextColor(YELLOW);
  panel->setCursor(6, CANVAS_H - 15);
  panel->printf("asking the AI (%s)...", MODES[mode].l1);

  size_t jpg_len = 0;
  uint8_t *jpg = canvasToJpeg(&jpg_len);
  if (!jpg) {
    showAnswer("Could not encode the drawing.");
    drawAnswerSide();
    led(255, 0, 0);
    return;
  }
  Serial.printf("drawing encoded: %u KB\n", (unsigned)(jpg_len / 1024));

  String answer;
  uint32_t t0 = millis();
  bool ok = askVision(jpg, jpg_len, MODES[mode].prompt, answer);
  uint32_t ms = millis() - t0;
  free(jpg);

  if (!ok) {
    showAnswer("No answer - see the serial monitor.");
    drawAnswerSide();
    led(255, 0, 0);
    return;
  }

  Serial.printf("AI (%lu ms): %s\n", (unsigned long)ms, answer.c_str());
  showAnswer(answer);
  drawAnswerSide();

#if SPEAK_REPLIES
  led(0, 255, 40);
  speak(answer);
#endif
  led(0, 0, 0);
}


/* ===========================================================================
 *  SETUP
 * =========================================================================== */
void setup() {
  Serial.begin(115200);
  delay(400);
  Serial.println("\n=== 08 Draw + AI  |  Robojax.com ===");

  pinMode(TFT_BLK, OUTPUT);
  digitalWrite(TFT_BLK, LOW);
  pinMode(SD_CS, OUTPUT);
  digitalWrite(SD_CS, HIGH);
  SPI.begin(TFT_SCLK, TFT_MISO, TFT_MOSI);

  panel->begin();
  panel->fillScreen(BLACK);
  digitalWrite(TFT_BLK, HIGH);

  if (!canvas->begin()) {
    panel->setTextColor(RED);
    panel->setTextSize(2);
    panel->setCursor(10, 100);
    panel->print("Canvas alloc failed!");
    panel->setTextSize(1);
    panel->setCursor(10, 130);
    panel->print("Tools > PSRAM > OPI PSRAM must be set.");
    while (1) delay(1000);
  }

  bbct.init(TOUCH_SDA, TOUCH_SCL, TOUCH_RST, TOUCH_INT);
  delay(50);

  rgb.begin();
  rgb.setBrightness(LED_BRIGHTNESS);
  led(0, 0, 0);

  spkInit();

  clearCanvas();
  panel->setTextSize(1);
  panel->setTextColor(YELLOW);
  panel->setCursor(BTN_X + 2, 4);
  panel->printf("WiFi...");

  WiFi.mode(WIFI_STA);
  WiFi.begin(WIFI_SSID, WIFI_PASS);
  uint32_t t0 = millis();
  while (WiFi.status() != WL_CONNECTED && millis() - t0 < 20000) delay(300);
  Serial.println(WiFi.status() == WL_CONNECTED
                   ? "WiFi ok" : "WiFi FAILED (2.4GHz only - check secrets.h)");

  drawButtons();
  Serial.println("Draw on the screen, then tap ASK. MODE cycles the question.");
}


/* ===========================================================================
 *  LOOP
 * =========================================================================== */
void loop() {
  /* edge-detected buttons: one action per tap, finger must lift first */
  static bool btn_down = false;
  static uint8_t misses = 0;

  uint16_t ux, uy;
  bool touched = getTouch(&ux, &uy);
  int x = touched ? (int)ux : -1;
  int y = touched ? (int)uy : -1;

  if (touched && x >= BTN_X) {
    misses = 0;
    if (!btn_down) {
      btn_down = true;
      last_x = last_y = -1;                      // never draw from a button tap

      if (y >= BTN_MODE_Y && y < BTN_MODE_Y + BTN_H) {
        if (showing_answer) {                    // BACK TO DRAW
          canvas->flush();                       // repaint the drawing
          drawButtons();
        } else {                                 // MODE
          mode = (mode + 1) % N_MODES;
          drawButtons();
          Serial.printf("mode: %s %s\n", MODES[mode].l1, MODES[mode].l2);
        }
      } else if (y >= BTN_ASK_Y && y < BTN_ASK_Y + BTN_H) {
        if (showing_answer) {
#if SPEAK_REPLIES
          led(0, 255, 40); playLastAnswer(); led(0, 0, 0);
#endif
        } else {
          askAboutDrawing();
        }
      } else if (y >= BTN_CLEAR_Y && y < BTN_CLEAR_Y + BTN_H) {
        clearCanvas();
        drawButtons();
      }
    }
  } else {
    if (btn_down && ++misses >= 4) { btn_down = false; misses = 0; }
    /* Draw only when no answer overlay is up and we are not mid-button-press. */
    if (!btn_down && !showing_answer) handleDrawing(touched, x, y);
  }

  /* The WiFi meter repaints only when the pen has been up for a moment, so it
   * can never steal SPI time in the middle of a stroke. */
  static uint32_t last_wifi = 0;
  if (last_x < 0 && millis() - last_wifi > 3000) { last_wifi = millis(); drawWifi(); }

  /* Poll as fast as possible while the pen is down - every millisecond spent
   * sleeping here is a touch sample missed and a gap in the line. */
  if (last_x < 0) delay(5);
}