Files
Seasoned/Seasoned.Backend/Services/RecipeService.cs
2026-03-12 20:17:25 +00:00

142 lines
5.5 KiB
C#

using Seasoned.Backend.DTOs;
using Mscc.GenerativeAI;
using Microsoft.AspNetCore.Http;
using System.IO;
using System.Text.Json;
using System.Text.Json.Serialization;
namespace Seasoned.Backend.Services;
public class RecipeService : IRecipeService
{
private readonly string _apiKey;
public RecipeService(IConfiguration config)
{
_apiKey = config["GEMINI_API_KEY"] ?? throw new ArgumentNullException("API Key missing");
}
public async Task<RecipeResponseDto> ParseRecipeImageAsync(IFormFile image)
{
var googleAI = new GoogleAI(_apiKey);
var model = googleAI.GenerativeModel("gemini-3.1-flash-lite-preview");
using var ms = new MemoryStream();
await image.CopyToAsync(ms);
var base64Image = Convert.ToBase64String(ms.ToArray());
var prompt = @"Extract the recipe details from this image.
RULES FOR THE 'icon' FIELD:
1. Select a valid 'Material Design Icon' name (e.g., 'mdi-pasta', 'mdi-bread-slice', 'mdi-muffin', 'mdi-pizza').
2. If the recipe type is ambiguous or you cannot find a specific matching icon, you MUST return 'mdi-silverware-fork-knife'.
IMPORTANT: Return ONLY a raw JSON string.
DO NOT include markdown formatting.
JSON structure:
{
""title"": ""string"",
""icon"": ""string"",
""ingredients"": [""string"", ""string""],
""instructions"": [""string"", ""string""]
}";
var generationConfig = new GenerationConfig {
ResponseMimeType = "application/json",
Temperature = 0.1f
};
var request = new GenerateContentRequest(prompt, generationConfig);
await Task.Run(() => request.AddMedia(base64Image, image.ContentType ?? "image/png"));
var response = await model.GenerateContent(request);
string rawText = response.Text?.Trim() ?? "";
int start = rawText.IndexOf('{');
int end = rawText.LastIndexOf('}');
if (start == -1 || end == -1)
{
return new RecipeResponseDto { Title = "Error" };
}
string cleanJson = rawText.Substring(start, (end - start) + 1);
try
{
var options = new JsonSerializerOptions { PropertyNameCaseInsensitive = true };
var result = JsonSerializer.Deserialize<RecipeResponseDto>(cleanJson, options);
if (result != null && string.IsNullOrEmpty(result.Icon))
{
result.Icon = "mdi-silverware-fork-knife";
}
return result ?? new RecipeResponseDto { Title = "Empty Response" };
}
catch (JsonException ex)
{
Console.WriteLine($"Raw AI Output: {rawText}");
Console.WriteLine($"Failed to parse JSON: {ex.Message}");
return new RecipeResponseDto { Title = "Parsing Error" };
}
}
public async Task<ChefConsultResponseDto> ConsultChefAsync(string userPrompt)
{
var googleAI = new GoogleAI(_apiKey);
var model = googleAI.GenerativeModel("gemini-3.1-flash-lite-preview");
var systemPrompt = @"You are the 'Seasoned' Head Chef, a master of real-world culinary arts.
You operate a professional kitchen and only provide advice that can be used in a real kitchen.
STRICT CONTENT RULES:
1. REAL FOOD ONLY: You specialize in real-world ingredients and techniques.
2. CELEBRITY CHEFS: You can provide recipes from real chefs like Gordon Ramsay or Julia Child.
3. FICTIONAL FOOD TRANSLATION: If a user asks for food from a game (like a Skyrim Sweetroll) or movie,
do NOT give game mechanics. Instead, provide a REAL-WORLD recipe that recreates that item.
In your 'reply', treat it like a fun culinary challenge.
4. REFUSAL POLICY: If the user asks about non-food topics (video game strategies, tech support, politics, or 'Minecraft crafting grids'),
politely refuse. Stay in character: 'The Chef's Grimoire is for spices, not spells' or 'I deal in pans, not pixels.'
RESPONSE FORMAT:
You MUST return ONLY a raw JSON object with these keys:
{
""reply"": ""A friendly, thematic response from the Chef."",
""recipe"": {
""title"": ""string"",
""icon"": ""string (must be a valid mdi- icon name)"",
""ingredients"": [""string"", ""string""],
""instructions"": [""string"", ""string""]
}
}
Note: Set the 'recipe' object to null if you are only chatting or refusing a non-food prompt.";
var fullPrompt = $"{systemPrompt}\n\nUser Question: {userPrompt}";
var generationConfig = new GenerationConfig {
ResponseMimeType = "application/json",
Temperature = 0.7f
};
var request = new GenerateContentRequest(fullPrompt, generationConfig);
var response = await model.GenerateContent(request);
try
{
var options = new JsonSerializerOptions { PropertyNameCaseInsensitive = true };
string jsonToParse = response.Text ?? "{ \"reply\": \"The chef is speechless. Try again?\" }";
var result = JsonSerializer.Deserialize<ChefConsultResponseDto>(jsonToParse, options);
return result ?? new ChefConsultResponseDto { Reply = "Chef is a bit confused!" };
}
catch (JsonException)
{
return new ChefConsultResponseDto { Reply = "The kitchen is a mess right now. Try again?" };
}
}
}