Add committment to value chain analysis

app_web.py

@@ -7,7 +7,7 @@ import streamlit as st
 from src.config_loader import load_config, get_api_key
 from src.ingest import get_chroma_collection, ingest_documents
 from src.kb_meta import load_kb_meta_brief
-from src.query_engine import understand_query, categorize_query, init_query
+from src.query_engine import understand_query, categorize_query, init_query, analyze_committment_to_vc
 from src.retriever import clear_collection_cache, retrieve
 from src.verifier import verify_and_respond
 from src.llm import list_models

src/prompts.py

@@ -315,10 +315,9 @@ Search for budget allocations, expenditure allocations, projects, programs, or p
 
 Identify:
 
-A. Budget allocation during FY2020/2021 (or nearest baseline year)
+A. Budget allocation at least five years prior to current date
 
-B. Budget allocation during FY2025/2026
-   (or most recent available year)
+B. Current budget allocation (or in the most recent financial year)
 
 Extract:
 
@@ -335,29 +334,32 @@ If multiple values are found, report all candidate values and explain difference
 
 STEP 2: STRATEGIC COMMITMENT
 
-Across all supplied policy documents determine:
+Extract all national agricultural strategies, national development plans, donor strategies, 
+and value chain specific strategies in the last 10 years.
+
+Across all supplied policy documents and identified strategies determine:
 
 1. Total number of documents examined
 
 2. Number of documents where the value chain is discussed
 
-3. Whether the value chain is discussed in:
+3. Whether:
 
-   a) Most recent National Development Plan
+   a) the value chain is identified as a priority 
+   in the national development plan
 
-   b) Most recent Agricultural Development Strategy,
-      Agricultural Sector Plan, Agricultural
-      Transformation Strategy, or equivalent
+   b) the value chain is identified as a priority 
+   in the most recent agricultural development strategy
 
-   c) Standalone value chain strategy
+   c) there is a standalone value chain strategy
 
-   d) Flagship value chain project
+   d) there is a flagship value chain project 
+   (supported by donors/government, or co-financed between them)
 
 For each item provide:
 
-- YES
-- NO
-- NOT_FOUND
+- YES (if value chain is only mentioned with explicit reference)
+- NO (if there is no mention or reference of the value chain for the respective category)
 
 and supporting evidence.
 
@@ -390,9 +392,8 @@ C. Dedicated ministry department,
 
 Return:
 
-YES
-NO
-NOT_FOUND
+- YES (if value chain is only mentioned with explicit reference)
+- NO (if there is no mention or reference of the value chain for the respective category)
 
 for each category.
 
@@ -413,15 +414,20 @@ Coverage Weight Formula:
 documents_discussing_value_chain /
 documents_examined
 
+CONTEXT (use ONLY this to answer)
+--------------------------------------------------------
+{context}
+--------------------------------------------------------
+
 --------------------------------------------------------
 
 OUTPUT FORMAT
 --------------------------------------------------------
 
-{
+{{
   "value_chain": "",
 
-  "financial_commitment": {
+  "financial_commitment": {{
     "budget_start_year": "",
     "budget_start_value": "",
     "budget_end_year": "",
@@ -429,66 +435,68 @@ OUTPUT FORMAT
     "currency": "",
     "responsible_institution": "",
     "evidence": [
-      {
+      {{
         "quote": "",
         "source_document": "",
         "page_or_section": ""
-      }
+      }}
     ]
-  },
+  }},
 
-  "strategic_commitment": {
+  "strategic_commitment": {{
     "documents_examined": 0,
     "documents_discussing_value_chain": 0,
     "coverage_weight": 0,
 
-    "development_plan": {
+    "development_plan": {{
       "value": "YES|NO|NOT_FOUND",
       "evidence": []
-    },
+    }},
 
-    "agricultural_strategy": {
+    "agricultural_strategy": {{
       "value": "YES|NO|NOT_FOUND",
       "evidence": []
-    },
+    }},
 
-    "standalone_vc_strategy": {
+    "standalone_vc_strategy": {{
       "value": "YES|NO|NOT_FOUND",
       "evidence": []
-    },
+    }},
 
-    "flagship_project": {
+    "flagship_project": {{
       "value": "YES|NO|NOT_FOUND",
       "evidence": []
-    }
-  },
+    }}
+  }},
 
-  "institutional_commitment": {
-    "dedicated_agency": {
+  "institutional_commitment": {{
+    "dedicated_agency": {{
       "value": "YES|NO|NOT_FOUND",
       "evidence": []
-    },
+    }},
 
-    "dedicated_research_institute": {
+    "dedicated_research_institute": {{
       "value": "YES|NO|NOT_FOUND",
       "evidence": []
-    },
+    }},
 
-    "dedicated_ministerial_department": {
+    "dedicated_ministerial_department": {{
       "value": "YES|NO|NOT_FOUND",
       "evidence": []
-    }
-  }
-}
+    }}
+  }}
+}}
 
 """
 
-def build_vc_prompt(value_chain:str,
-                    domain:str
-                   ):
+def build_vc_analyzer_prompt(context:str,
+                             value_chain:str,
+                             domain:str
+                             ):
     return VALUE_CHAIN_PROMPT_TEMPLATE.format(
         domain=domain,
         value_chain=value_chain,
+        context=context
     )
 
 def dict_to_string(d, indent=0):

src/query_engine.py

@@ -6,7 +6,87 @@ import re
 from typing import Dict
 
 from src.llm import generate
-from src.prompts import build_query_understanding_prompt, build_categorizer_prompt, build_init_prompt
+from src.prompts import (build_query_understanding_prompt, 
+                         build_categorizer_prompt, build_init_prompt, build_vc_analyzer_prompt)
+
+def analyze_committment_to_vc(value_chain:str,
+                              cfg:Dict,):
+    domain = cfg.get("chatbot", {}).get("domain", "research")
+    
+    prompt = build_vc_analyzer_prompt(value_chain, domain)
+
+    try:
+        raw = generate(
+            "",
+            prompt,
+            cfg,
+            max_tokens=512,
+        )
+        return _parse_vc_analyzer_result(raw)
+    except Exception:
+        # LLM failure: fall back to raw query
+        return {
+            "value_chain": None,
+
+            "financial_commitment": {
+                "budget_start_year": "",
+                "budget_start_value": "",
+                "budget_end_year": "",
+                "budget_end_value": "",
+                "currency": "",
+                "responsible_institution": "",
+                "evidence": [
+                {
+                    "quote": "",
+                    "source_document": "",
+                    "page_or_section": ""
+                }
+                ]
+            },
+
+            "strategic_commitment": {
+                "documents_examined": 0,
+                "documents_discussing_value_chain": 0,
+                "coverage_weight": 0,
+
+                "development_plan": {
+                "value": "NOT_FOUND",
+                "evidence": []
+                },
+
+                "agricultural_strategy": {
+                "value": "NOT_FOUND",
+                "evidence": []
+                },
+
+                "standalone_vc_strategy": {
+                "value": "NOT_FOUND",
+                "evidence": []
+                },
+
+                "flagship_project": {
+                "value": "NOT_FOUND",
+                "evidence": []
+                }
+            },
+
+            "institutional_commitment": {
+                "dedicated_agency": {
+                "value": "NOT_FOUND",
+                "evidence": []
+                },
+
+                "dedicated_research_institute": {
+                "value": "NOT_FOUND",
+                "evidence": []
+                },
+
+                "dedicated_ministerial_department": {
+                "value": "NOT_FOUND",
+                "evidence": []
+                }
+            }
+        }
 
 
 def _load_sql_schema_summary(sql_db_dir: str) -> str:
@@ -168,6 +248,93 @@ def understand_query(
 
     return result
 
+def _parse_vc_analyzer_result(raw:str) -> Dict:
+    parsed = None
+
+    # Try direct parse
+    try:
+        parsed = json.loads(raw)
+    except (json.JSONDecodeError, TypeError):
+        pass
+
+    # Fallback: scan for the first valid JSON object using raw_decode
+    if parsed is None:
+        decoder = json.JSONDecoder()
+        for i, ch in enumerate(raw):
+            if ch == '{':
+                try:
+                    parsed, _ = decoder.raw_decode(raw, i)
+                    break
+                except json.JSONDecodeError:
+                    continue
+
+    # Unparseable: fall back to raw query
+    if not isinstance(parsed, dict):
+        return {
+            "value_chain": None,
+
+            "financial_commitment": {
+                "budget_start_year": "",
+                "budget_start_value": "",
+                "budget_end_year": "",
+                "budget_end_value": "",
+                "currency": "",
+                "responsible_institution": "",
+                "evidence": [
+                {
+                    "quote": "",
+                    "source_document": "",
+                    "page_or_section": ""
+                }
+                ]
+            },
+
+            "strategic_commitment": {
+                "documents_examined": 0,
+                "documents_discussing_value_chain": 0,
+                "coverage_weight": 0,
+
+                "development_plan": {
+                "value": "NOT_FOUND",
+                "evidence": []
+                },
+
+                "agricultural_strategy": {
+                "value": "NOT_FOUND",
+                "evidence": []
+                },
+
+                "standalone_vc_strategy": {
+                "value": "NOT_FOUND",
+                "evidence": []
+                },
+
+                "flagship_project": {
+                "value": "NOT_FOUND",
+                "evidence": []
+                }
+            },
+
+            "institutional_commitment": {
+                "dedicated_agency": {
+                "value": "NOT_FOUND",
+                "evidence": []
+                },
+
+                "dedicated_research_institute": {
+                "value": "NOT_FOUND",
+                "evidence": []
+                },
+
+                "dedicated_ministerial_department": {
+                "value": "NOT_FOUND",
+                "evidence": []
+                }
+            }
+        }
+
+    return parsed
+
 def _parse_qinit_result(raw:str) -> Dict:
     parsed = None
 
@@ -197,6 +364,7 @@ def _parse_qinit_result(raw:str) -> Dict:
         }
 
     return parsed
+
 def _parse_qcat_result(raw:str) -> Dict:
     parsed = None