BUG fixes buerokratt#192 buerokratt#189 (buerokratt#211)

* updated docker compose ec2

* integrate streaming endpoint with test prodction connection page

* formatted response with markdown

* fe logic for the encryption

* vault secret update after fixing issues

* fixed formatting issue

* integration with be

* update cron manager vault script

* tested integration of vault security update

* fix security issues

* fixed issue references are not sending with streming tokens

* complete buerokratt#192 and buerokratt#206 bug fixes

* change production inference display logic

* Remove obsolete Vite configuration files and associated plugins

* added new model

---------

Co-authored-by: Thiru Dinesh <56014038+Thirunayan22@users.noreply.github.com>
Co-authored-by: Thiru Dinesh <thiru.dinesh@rootcodelabs.com>
Co-authored-by: erangi-ar <erangika.ariyasena@rootcode.io>

Author: 3 people

DSL/Liquibase/changelog/rag-search-script-v1-llm-connections.sql

@@ -102,6 +102,7 @@ INSERT INTO llm_models (platform_id, model_key, model_name) VALUES
 -- Azure models
 ((SELECT id FROM llm_platforms WHERE platform_key = 'azure'), 'gpt-4o-mini', 'GPT-4o-mini'),
 ((SELECT id FROM llm_platforms WHERE platform_key = 'azure'), 'gpt-4o', 'GPT-4o'),
+((SELECT id FROM llm_platforms WHERE platform_key = 'azure'), 'gpt-4.1', 'GPT-4.1'),
 -- AWS models
 ((SELECT id FROM llm_platforms WHERE platform_key = 'aws'), 'anthropic-claude-3.5-sonnet', 'Anthropic Claude 3.5 Sonnet'),
 ((SELECT id FROM llm_platforms WHERE platform_key = 'aws'), 'anthropic-claude-3.7-sonnet', 'Anthropic Claude 3.7 Sonnet');

README.md

@@ -12,7 +12,7 @@ The **BYK-RAG Module** is part of the Burokratt ecosystem, designed to provide *
   - Models searchable via dropdown with cache-enabled indicators.
 
 - **Enhanced Security with RSA Encryption**  
-  - LLM credentials encrypted with RSA-2048 asymmetric encryption before storage.  
+  - LLM credentials encrypted with RSA-2048 asymmetric encryption before storage.
   - GUI encrypts using public key; CronManager decrypts with private key.  
   - Additional security layer beyond HashiCorp Vault's encryption.  
 

src/llm_orchestration_service.py

@@ -32,6 +32,13 @@ llm:
           temperature: 0.5
           deployment_name: "gpt-4o-deployment"
 
+        gpt-4.1:
+          model_type: "chat"
+          max_tokens: 13107
+          temperature: 0.6
+          deployment_name: "gpt-4.1"
+
+
     # AWS Bedrock Configuration
     aws_bedrock:
       cache: true  # Keep caching enabled (DSPY default)

src/llm_orchestrator_config/config/llm_config.yaml

@@ -1,20 +1,36 @@
-OUT_OF_SCOPE_MESSAGE = (
-    "I apologize, but I’m unable to provide a complete response because the available "
-    "context does not sufficiently cover your request. Please try rephrasing or providing more details."
-)
-
-TECHNICAL_ISSUE_MESSAGE = (
-    "Technical issue with response generation\n"
-    "I apologize, but I’m currently unable to generate a response due to a temporary technical issue. "
-    "Please try again in a moment."
-)
+# Multilingual message dictionaries
+OUT_OF_SCOPE_MESSAGES = {
+    "et": "Vabandust, kuid mul pole piisavalt konteksti, et teie küsimusele vastata. Palun püüdke ümber sõnastada või lisage rohkem üksikasju.",
+    "ru": "Извините, но у меня недостаточно контекста для ответа на ваш вопрос. Пожалуйста, попробуйте переформулировать или предоставить больше деталей.",
+    "en": "I apologize, but I'm unable to provide a complete response because the available context does not sufficiently cover your request. Please try rephrasing or providing more details.",
+}
+
+TECHNICAL_ISSUE_MESSAGES = {
+    "et": "Tehniline probleem vastuse genereerimisel\nVabandust, kuid ma ei saa praegu vastust genereerida ajutise tehnilise probleemi tõttu. Palun proovige mõne hetke pärast uuesti.",
+    "ru": "Техническая проблема при генерации ответа\nИзвините, в настоящее время я не могу сгенерировать ответ из-за временной технической проблемы. Пожалуйста, попробуйте еще раз через мгновение.",
+    "en": "Technical issue with response generation\nI apologize, but I'm currently unable to generate a response due to a temporary technical issue. Please try again in a moment.",
+}
+
+INPUT_GUARDRAIL_VIOLATION_MESSAGES = {
+    "et": "Vabandust, kuid ma ei saa selle taotlusega aidata, kuna see rikub meie kasutustingimusi.",
+    "ru": "Извините, но я не могу помочь с этим запросом, так как он нарушает нашу политику использования.",
+    "en": "I apologize, but I'm unable to assist with that request as it violates our usage policies.",
+}
+
+OUTPUT_GUARDRAIL_VIOLATION_MESSAGES = {
+    "et": "Vabandust, kuid ma ei saa vastust anda, kuna see võib rikkuda meie kasutustingimusi.",
+    "ru": "Извините, но я не могу предоставить ответ, так как он может нарушить нашу политику использования.",
+    "en": "I apologize, but I'm unable to provide a response as it may violate our usage policies.",
+}
+
+# Legacy constants for backward compatibility (English defaults)
+OUT_OF_SCOPE_MESSAGE = OUT_OF_SCOPE_MESSAGES["en"]
+TECHNICAL_ISSUE_MESSAGE = TECHNICAL_ISSUE_MESSAGES["en"]
+INPUT_GUARDRAIL_VIOLATION_MESSAGE = INPUT_GUARDRAIL_VIOLATION_MESSAGES["en"]
+OUTPUT_GUARDRAIL_VIOLATION_MESSAGE = OUTPUT_GUARDRAIL_VIOLATION_MESSAGES["en"]
 
 UNKNOWN_SOURCE = "Unknown source"
 
-INPUT_GUARDRAIL_VIOLATION_MESSAGE = "I apologize, but I'm unable to assist with that request as it violates our usage policies."
-
-OUTPUT_GUARDRAIL_VIOLATION_MESSAGE = "I apologize, but I'm unable to provide a response as it may violate our usage policies."
-
 GUARDRAILS_BLOCKED_PHRASES = [
     "i'm sorry, i can't respond to that",
     "i cannot respond to that",
@@ -88,6 +104,22 @@
 
 VALIDATION_GENERIC_ERROR = "I apologize, but I couldn't process your request. Please check your input and try again."
 
+
+# Helper function to get localized messages
+def get_localized_message(message_dict: dict, language_code: str = "en") -> str:
+    """
+    Get message in the specified language, fallback to English.
+
+    Args:
+        message_dict: Dictionary with language codes as keys
+        language_code: Language code ('et', 'ru', 'en')
+
+    Returns:
+        Localized message string
+    """
+    return message_dict.get(language_code, message_dict.get("en", ""))
+
+
 # Service endpoints
 RAG_SEARCH_RESQL = "http://resql:8082/rag-search"
 RAG_SEARCH_RUUTER_PUBLIC = "http://ruuter-public:8086/rag-search"

src/llm_orchestrator_config/llm_ochestrator_constants.py

@@ -27,6 +27,13 @@ class ConversationHistory(BaseModel):
 class PromptRefiner(dspy.Signature):
     """Produce N distinct, concise rewrites of the user's question using chat history.
 
+    CRITICAL LANGUAGE RULE:
+    - The rewrites MUST be in the SAME language as the input question
+    - Estonian question → Estonian rewrites
+    - Russian question → Russian rewrites
+    - English question → English rewrites
+    - Preserve the natural language of the original question
+
     Constraints:
     - Preserve the original intent; don't inject unsupported constraints.
     - Resolve pronouns with context when safe; avoid changing semantics.
@@ -36,11 +43,13 @@ class PromptRefiner(dspy.Signature):
     """
 
     history: str = dspy.InputField(desc="Recent conversation history (turns).")
-    question: str = dspy.InputField(desc="The user's latest question to refine.")
+    question: str = dspy.InputField(
+        desc="The user's latest question to refine. Preserve its language in rewrites."
+    )
     n: int = dspy.InputField(desc="Number of rewrites to produce (N).")
 
     rewrites: list[str] = dspy.OutputField(
-        desc="Exactly N refined variations of the question, each a single sentence."
+        desc="Exactly N refined variations of the question in THE SAME LANGUAGE as input, each a single sentence."
     )
 
 

src/prompt_refine_manager/prompt_refiner.py

@@ -22,16 +22,27 @@
 class ResponseGenerator(dspy.Signature):
     """Produce a grounded answer from the provided context ONLY.
 
+    CRITICAL LANGUAGE RULE:
+    - The answer MUST be in the SAME language as the input question
+    - Estonian question → Estonian answer
+    - Russian question → Russian answer
+    - English question → English answer
+    - Maintain the natural language flow and grammar of the detected language
+
     Rules:
     - Use ONLY the provided context blocks; do not invent facts.
     - If the context is insufficient, set questionOutOfLLMScope=true and say so briefly.
     - Do not include citations in the 'answer' field.
     """
 
-    question: str = dspy.InputField()
+    question: str = dspy.InputField(
+        desc="User's question. Answer in the SAME language as this question."
+    )
     context_blocks: List[str] = dspy.InputField()
     citations: List[str] = dspy.InputField()
-    answer: str = dspy.OutputField(desc="Human-friendly answer without citations")
+    answer: str = dspy.OutputField(
+        desc="Human-friendly answer in THE SAME LANGUAGE as the question, without citations"
+    )
     questionOutOfLLMScope: bool = dspy.OutputField(
         desc="True if context is insufficient to answer"
     )
@@ -40,6 +51,8 @@ class ResponseGenerator(dspy.Signature):
 class ScopeChecker(dspy.Signature):
     """Quick check if question can be answered from context.
 
+    LANGUAGE NOTE: This is an internal check, language doesn't matter for scope determination.
+
     Rules:
     - Return True ONLY if context is completely insufficient
     - Return False if context has ANY relevant information

src/response_generator/response_generate.py

@@ -0,0 +1,116 @@
+"""Language detection utility for multilingual support.
+
+Detects Estonian, Russian, and English based on character patterns and common words.
+"""
+
+import re
+from typing import Literal
+from loguru import logger
+
+LanguageCode = Literal["et", "ru", "en"]
+
+
+def detect_language(text: str) -> LanguageCode:
+    """
+    Detect language from input text.
+
+    Detection Strategy:
+    1. Check for Cyrillic characters (Russian)
+    2. Check for Estonian-specific characters
+    3. Check for Estonian common words
+    4. Default to English
+
+    Args:
+        text: Input text to analyze
+
+    Returns:
+        Language code: 'et' (Estonian), 'ru' (Russian), 'en' (English)
+
+    Examples:
+        >>> detect_language("Mis on sünnitoetus?")
+        'et'
+        >>> detect_language("Что такое пособие?")
+        'ru'
+        >>> detect_language("What is the benefit?")
+        'en'
+    """
+    if not text or not text.strip():
+        logger.warning(
+            "Empty text provided for language detection, defaulting to English"
+        )
+        return "en"
+
+    text_sample = text.strip()[:500]  # Use first 500 chars for detection
+
+    # Check for Cyrillic characters (Russian) - use percentage-based detection
+    cyrillic_count = len(re.findall(r"[а-яА-ЯёЁ]", text_sample))
+    total_alpha = len(re.findall(r"[a-zA-Zа-яА-ЯёЁõäöüšžÕÄÖÜŠŽ]", text_sample))
+
+    if (
+        total_alpha > 0 and cyrillic_count / total_alpha > 0.25
+    ):  # 25% Cyrillic threshold
+        logger.debug(
+            f"Detected Russian (Cyrillic: {cyrillic_count}/{total_alpha} = {cyrillic_count / total_alpha:.1%})"
+        )
+        return "ru"
+
+    # Check for Estonian-specific characters (õ, ä, ö, ü, š, ž)
+    estonian_chars = re.findall(r"[õäöüšž]", text_sample, re.IGNORECASE)
+    if len(estonian_chars) > 0:
+        logger.debug(f"Detected Estonian (special chars: {len(estonian_chars)})")
+        return "et"
+
+    # Check for Estonian common words - use distinctive markers to avoid English false positives
+    estonian_markers = [
+        "kuidas",
+        "miks",
+        "kus",
+        "millal",
+        "kes",
+        "võib",
+        "olen",
+        "oled",
+        "see",
+        "seda",
+        "jah",
+        "või",
+        "ning",
+        "siis",
+        "veel",
+        "aga",
+        "kuid",
+        "nii",
+        "nagu",
+        "oli",
+        "mis",
+    ]
+
+    # Tokenize and check for Estonian markers
+    words = re.findall(r"\b\w+\b", text_sample.lower())
+    estonian_word_count = sum(1 for word in words if word in estonian_markers)
+
+    # Scale threshold based on text length for better accuracy
+    threshold = 1 if len(words) < 10 else 2
+    if estonian_word_count >= threshold:
+        logger.debug(
+            f"Detected Estonian (marker words: {estonian_word_count}/{len(words)}, threshold: {threshold})"
+        )
+        return "et"
+
+    # Default to English
+    logger.debug("Detected English (default)")
+    return "en"
+
+
+def get_language_name(language_code: LanguageCode) -> str:
+    """
+    Get human-readable language name from code.
+
+    Args:
+        language_code: ISO 639-1 language code
+
+    Returns:
+        Full language name
+    """
+    language_names = {"et": "Estonian", "ru": "Russian", "en": "English"}
+    return language_names.get(language_code, "Unknown")