MarketAlly.AIPlugin.Extensions/MarketAlly.AIPlugin.Context/Search/SemanticSearchEnhancer.cs

using System.Net.Http.Headers;
using System.Text;
using System.Text.Json;
using Microsoft.Extensions.Logging;
using MarketAlly.AIPlugin.Context.Configuration;

namespace MarketAlly.AIPlugin.Context.Search
{
	/// <summary>
	/// Provides semantic search capabilities using OpenAI embeddings
	/// </summary>
	public class SemanticSearchEnhancer
	{
		private readonly ContextConfiguration _configuration;
		private readonly ILogger<SemanticSearchEnhancer> _logger;
		private readonly HttpClient _httpClient;
		private readonly Dictionary<string, float[]> _embeddingCache;
		private readonly SemaphoreSlim _rateLimitSemaphore;

		public SemanticSearchEnhancer(ContextConfiguration configuration, ILogger<SemanticSearchEnhancer> logger, HttpClient httpClient)
		{
			_configuration = configuration;
			_logger = logger;
			_httpClient = httpClient;
			_embeddingCache = new Dictionary<string, float[]>();
			_rateLimitSemaphore = new SemaphoreSlim(5, 5); // Limit concurrent API calls

			ConfigureHttpClient();
		}

		/// <summary>
		/// Calculates semantic similarity between query and content using embeddings
		/// </summary>
		public async Task<double> CalculateSemanticSimilarityAsync(string query, string content, CancellationToken cancellationToken = default)
		{
			if (string.IsNullOrEmpty(_configuration.Search.OpenAIApiKey))
			{
				_logger.LogWarning("OpenAI API key not configured, semantic search disabled");
				return 0.0;
			}

			try
			{
				await _rateLimitSemaphore.WaitAsync(cancellationToken);
				
				var queryEmbedding = await GetEmbeddingAsync(query, cancellationToken);
				var contentEmbedding = await GetEmbeddingAsync(content, cancellationToken);

				if (queryEmbedding != null && contentEmbedding != null)
				{
					var similarity = CalculateCosineSimilarity(queryEmbedding, contentEmbedding);
					_logger.LogDebug("Calculated semantic similarity: {Similarity} for query length {QueryLength} and content length {ContentLength}",
						similarity, query.Length, content.Length);
					return similarity;
				}

				return 0.0;
			}
			catch (Exception ex)
			{
				_logger.LogError(ex, "Failed to calculate semantic similarity");
				return 0.0;
			}
			finally
			{
				_rateLimitSemaphore.Release();
			}
		}

		/// <summary>
		/// Gets embedding for text, with caching to reduce API calls
		/// </summary>
		private async Task<float[]?> GetEmbeddingAsync(string text, CancellationToken cancellationToken)
		{
			// Truncate very long text to avoid API limits
			if (text.Length > 8000)
			{
				text = text.Substring(0, 8000);
			}

			// Check cache first
			var cacheKey = GenerateCacheKey(text);
			if (_embeddingCache.TryGetValue(cacheKey, out var cachedEmbedding))
			{
				return cachedEmbedding;
			}

			try
			{
				var requestBody = new
				{
					input = text,
					model = _configuration.Search.OpenAIEmbeddingModel
				};

				var jsonContent = JsonSerializer.Serialize(requestBody);
				var content = new StringContent(jsonContent, Encoding.UTF8, "application/json");

				var response = await _httpClient.PostAsync("https://api.openai.com/v1/embeddings", content, cancellationToken);
				
				if (!response.IsSuccessStatusCode)
				{
					var errorContent = await response.Content.ReadAsStringAsync(cancellationToken);
					_logger.LogError("OpenAI API error: {StatusCode} - {Error}", response.StatusCode, errorContent);
					return null;
				}

				var responseJson = await response.Content.ReadAsStringAsync(cancellationToken);
				var embeddingResponse = JsonSerializer.Deserialize<OpenAIEmbeddingResponse>(responseJson);

				if (embeddingResponse?.Data?.Length > 0)
				{
					var embedding = embeddingResponse.Data[0].Embedding;
					
					// Cache the embedding (with some memory management)
					if (_embeddingCache.Count > 1000) // Limit cache size
					{
						var oldestKey = _embeddingCache.Keys.First();
						_embeddingCache.Remove(oldestKey);
					}
					
					_embeddingCache[cacheKey] = embedding;
					return embedding;
				}

				return null;
			}
			catch (HttpRequestException ex)
			{
				_logger.LogError(ex, "Network error while getting embedding");
				return null;
			}
			catch (JsonException ex)
			{
				_logger.LogError(ex, "JSON parsing error while processing embedding response");
				return null;
			}
		}

		/// <summary>
		/// Calculates cosine similarity between two embedding vectors
		/// </summary>
		private double CalculateCosineSimilarity(float[] vectorA, float[] vectorB)
		{
			if (vectorA.Length != vectorB.Length)
			{
				_logger.LogWarning("Vector length mismatch: {LengthA} vs {LengthB}", vectorA.Length, vectorB.Length);
				return 0.0;
			}

			double dotProduct = 0.0;
			double magnitudeA = 0.0;
			double magnitudeB = 0.0;

			for (int i = 0; i < vectorA.Length; i++)
			{
				dotProduct += vectorA[i] * vectorB[i];
				magnitudeA += vectorA[i] * vectorA[i];
				magnitudeB += vectorB[i] * vectorB[i];
			}

			magnitudeA = Math.Sqrt(magnitudeA);
			magnitudeB = Math.Sqrt(magnitudeB);

			if (magnitudeA == 0.0 || magnitudeB == 0.0)
			{
				return 0.0;
			}

			var similarity = dotProduct / (magnitudeA * magnitudeB);
			
			// Normalize to 0-1 range (cosine similarity can be -1 to 1)
			return (similarity + 1.0) / 2.0;
		}

		/// <summary>
		/// Generates a cache key for text content
		/// </summary>
		private string GenerateCacheKey(string text)
		{
			// Use a simple hash for caching
			return text.Length > 100 ? 
				$"{text.Substring(0, 50)}_{text.GetHashCode()}_{text.Length}" : 
				text.GetHashCode().ToString();
		}

		/// <summary>
		/// Configures the HTTP client for OpenAI API calls
		/// </summary>
		private void ConfigureHttpClient()
		{
			if (!string.IsNullOrEmpty(_configuration.Search.OpenAIApiKey))
			{
				_httpClient.DefaultRequestHeaders.Authorization = 
					new AuthenticationHeaderValue("Bearer", _configuration.Search.OpenAIApiKey);
			}

			_httpClient.DefaultRequestHeaders.Add("User-Agent", "MarketAlly-Context-Plugin/1.0");
			_httpClient.Timeout = TimeSpan.FromSeconds(30);
		}

		/// <summary>
		/// Gets embedding cache statistics
		/// </summary>
		public EmbeddingCacheStats GetCacheStats()
		{
			return new EmbeddingCacheStats
			{
				CachedEmbeddings = _embeddingCache.Count,
				CacheHitRatio = 0.0, // Would need to track hits/misses for this
				IsEnabled = !string.IsNullOrEmpty(_configuration.Search.OpenAIApiKey)
			};
		}

		/// <summary>
		/// Clears the embedding cache
		/// </summary>
		public void ClearCache()
		{
			_embeddingCache.Clear();
			_logger.LogInformation("Embedding cache cleared");
		}
	}

	/// <summary>
	/// Response structure for OpenAI embeddings API
	/// </summary>
	public class OpenAIEmbeddingResponse
	{
		public OpenAIEmbeddingData[] Data { get; set; } = Array.Empty<OpenAIEmbeddingData>();
	}

	public class OpenAIEmbeddingData
	{
		public float[] Embedding { get; set; } = Array.Empty<float>();
	}

	/// <summary>
	/// Statistics for embedding cache performance
	/// </summary>
	public class EmbeddingCacheStats
	{
		public int CachedEmbeddings { get; set; }
		public double CacheHitRatio { get; set; }
		public bool IsEnabled { get; set; }
	}
}