[186] Implement ai multi users

.gitignore

@@ -2,6 +2,8 @@ node_modules
 CLAUDE.md
 .vercel
 .env
+key.pem
+*.pem
 
 .vscode/
 .github/agents
@@ -11,3 +13,14 @@ recordings/
 
 # Local design specs and implementation plans (not pushed)
 docs/
+
+# AI build artifacts and local dev files
+ai/core/server/build/
+ai/**/.venv/
+ai/**/venv/
+ai/llm-core/models/
+ai/llm-core/sup-it-lora-v3/
+ai/llm-core/unsloth_compiled_cache/
+ai/microservices/**/llm/
+*.wav
+*.mp4

ai/core/server/inc/network/MicroservicesManager.hpp

@@ -27,7 +27,8 @@
 #include <condition_variable>
 #include <queue>
 #include <memory>
-#include <crow.h>
+#include <functional>
+#include <future>
 
 #include "ExceptionManager.hpp"
 
@@ -74,6 +75,14 @@ namespace talkup_network {
              */
             static void send_to_sts_microservice(const nlohmann::json &data, ResponseCallback callback);
 
+            /**
+             * @brief Push structured simulation context (company, job offer) to STS for one interview.
+             * @return true if STS acknowledged registration.
+             */
+            static bool send_simulation_context_to_sts(
+                const std::string &interview_id,
+                const nlohmann::json &context_data);
+
             /**
              * @brief Initialize WebSocket connections to all registered microservices.
              * It's establishes persistent WebSocket connections to each microservice
@@ -108,9 +117,18 @@ namespace talkup_network {
         protected:
         private:
             struct WebSocketConnection {
+                enum class StsJobKind {
+                    StreamChunk,
+                    SimulationContext,
+                };
+
                 struct StsJob {
+                    StsJobKind kind = StsJobKind::StreamChunk;
                     nlohmann::json data;
                     ResponseCallback callback;
+                    std::string interview_id;
+                    nlohmann::json context_data;
+                    std::shared_ptr<std::promise<bool>> context_promise;
                 };
 
                 std::shared_ptr<boost::asio::io_context> io_context;
@@ -168,5 +186,10 @@ namespace talkup_network {
              * @param data The JSON data containing the job information.
              */
             static void process_sts_job(const nlohmann::json &data, ResponseCallback callback);
+
+            static void process_simulation_context_job(
+                const std::string &interview_id,
+                const nlohmann::json &context_data,
+                const std::shared_ptr<std::promise<bool>> &result_promise);
     };
 }

ai/core/server/inc/network/WebsocketManager.hpp

@@ -86,6 +86,9 @@ namespace talkup_network {
             void handle_stream_chunk(const nlohmann::json& json, crow::websocket::connection& conn,
                 std::shared_ptr<MicroservicesManager> microservices_manager);
 
+            void handle_simulation_context(const nlohmann::json& json, crow::websocket::connection& conn,
+                std::shared_ptr<MicroservicesManager> microservices_manager);
+
         private:
             std::unordered_map<std::string, std::function<void(const nlohmann::json&,
                 crow::websocket::connection&, std::shared_ptr<MicroservicesManager>)>> _type_handlers;

ai/core/server/protocol/EN/Communication_Protocol_AI_Server_Frontend_EN.md

@@ -84,6 +84,17 @@ Type format can be `audio`, `video`, `image` or `text` depending on the stream c
 ```
 
 ---
+
+## 4.5 Simulation context — company and job offer
+
+Before audio `stream_chunk` messages, send `simulation_context` so the STS LLM receives company/job context in its system prompt.
+
+**Order:** WebSocket open → `simulation_context` → `simulation_context_ack` → `stream_chunk`.
+
+See the French protocol document for the full JSON schema (`company`, `jobOffer`, `additionalInfo`, `language`, `interviewType`).
+
+---
+
 ## 5. Advanced Level — AI Server ↔ Microservices
 The protocol is designed to evolve into a modular architecture where the AI Server delegates tasks to Python microservices.
 

ai/core/server/protocol/FR/Protocole_Communication_ServeurIA_Frontend_FR.md

@@ -85,6 +85,53 @@ Le format de type peut être `audio`, `video`, `image` ou `text` selon le conten
 
 ---
 
+## 4.5 Contexte de simulation — entreprise et offre d'emploi
+
+Avant d'envoyer des `stream_chunk` audio, le Frontend **doit** enregistrer le contexte métier via `simulation_context`. Ce contexte alimente le prompt système du LLM (STS).
+
+**Ordre recommandé :** connexion WS → `simulation_context` → `simulation_context_ack` → `stream_chunk`.
+
+### Exemple : Frontend → Serveur IA
+
+```json
+{
+  "key": "exemple_key",
+  "type": "simulation_context",
+  "stream_id": "019bf2a0-....",
+  "format": "json",
+  "timestamp": 1739592334,
+  "data": {
+    "language": "French",
+    "interviewType": "technical",
+    "company": {
+      "name": "Acme Digital",
+      "description": "ESN specialisee en transformation digitale."
+    },
+    "jobOffer": {
+      "title": "Developpeur Backend Java",
+      "description": "Mission SI bancaire microservices.",
+      "requirements": "Java 17, Spring Boot, Kafka."
+    },
+    "additionalInfo": "Notes complementaires (CV resume, objectifs)."
+  }
+}
+```
+
+### Exemple : Serveur IA → Frontend
+
+```json
+{
+  "key": "exemple_key",
+  "type": "simulation_context_ack",
+  "stream_id": "019bf2a0-....",
+  "format": "text",
+  "timestamp": 1739592335,
+  "data": "simulation context registered"
+}
+```
+
+---
+
 ## 5. Niveau avancé — Serveur IA ↔ Microservices
 Le protocole est conçu pour évoluer vers une architecture modulaire où le Serveur IA délègue certaines tâches à des microservices Python.