perf: optimize token usage and context window for K2 engine (10k context, 8k response)

app/services/analysis_service.py

@@ -102,12 +102,13 @@ class AnalysisService:
                         
                         if remote_data:
                             p_info = remote_data[0]
+                            snippet = p_info.get("content", p_info.get("summary", ""))[:10000]
                             paper = ResearchPaper(
                                 project_id=project_id,
                                 remote_id=pid,
                                 title=p_info.get("title", "Unknown"),
                                 authors=", ".join(p_info.get("authors", [])),
-                                summary=p_info.get("summary", ""),
+                                summary=snippet,
                                 publication_year=datetime.now().year # Fallback
                             )
                             db.add(paper)
@@ -135,7 +136,7 @@ class AnalysisService:
                         title=paper.title,
                         authors=author_list,
                         abstract=paper.summary or "",
-                        content=paper.summary or "", 
+                        content=(paper.summary or "")[:10000], 
                         document_type=dtype,
                         url=paper.pdf_path or ""
                     ))

app/services/k2_think_engine.py

@@ -62,14 +62,14 @@ class K2ThinkEngine:
                 openai_api_key=settings.K2_THINK_API_KEY,
                 openai_api_base=settings.K2_THINK_API_URL,
                 temperature=0.7,
-                max_tokens=4096,
+                max_tokens=8192,
                 max_retries=3 
             )
 
             # 2. Préparation du contexte documentaire
             context_parts = []
             for doc in request.documents:
-                snippet = doc.content[:30000]
+                snippet = doc.content[:10000]
                 first_author = doc.authors[0].split()[-1] if doc.authors else "Unknown"
                 year = "n.d."
                 citation_key = f"({first_author}, {year})"
@@ -87,6 +87,7 @@ class K2ThinkEngine:
             # 4. Définition du Prompt Système
             system_template = """You are the K2 Think V2 Scientific Co-Investigator.
 Your primary directive is MULTI-DOCUMENT REASONING and KNOWLEDGE SYNTHESIS.
+Keep your reasoning process efficient and focused. Proceed to the [RESULT] JSON block as soon as you have synthesized the core findings.
 
 IMPORTANT: Your internal reasoning (thoughts) should be contained within <think></think> tags.