update

src/interface.jl

@@ -127,7 +127,7 @@ function decisionMaker(a::T1, state::T2)::Dict{Symbol, Any} where {T1<:agent, T2
   Thought can reason about the current situation, and Action can be three types:
   1) winestock[query], which you can use to find wine in your inventory. The more input data the better.
   2) chatbox[text], which you can use to interact with the user.
-  3) recommendation[answer], which returns your wine reccommendation to the user. 
+  3) reccommendbox[answer], which returns your wine reccommendation to the user. 
   
   $responseformat
   
@@ -186,9 +186,18 @@ function decisionMaker(a::T1, state::T2)::Dict{Symbol, Any} where {T1<:agent, T2
 
   _response = GeneralUtils.sendReceiveMqttMsg(outgoingMsg)
   _thoughtJsonStr = _response[:response][:text]
-  thoughtJsonStr = jsoncorrection(a, _thoughtJsonStr, responseformat)
+  expectedJsonExample =
+  """
+  Here is an expected JSON format:
+  {
+    "Thought": "...",
+    "Action": {"name": "...", "input": "..."},
+    "Observation": "..."
+  }
+  """
+  thoughtJsonStr = jsoncorrection(a, _thoughtJsonStr, expectedJsonExample)
   thoughtDict = copy(JSON3.read(thoughtJsonStr)) 
-  pprint(thoughtDict)
+
   return thoughtDict
 end
 
@@ -219,10 +228,7 @@ function progressValueEstimator(a::T1, state::T2)::Tuple{String, Integer} where
   """
   You should only respond in JSON format as describe below:
   {
-    "Thought_1": "reasoning 1",
-    "Action_1": {"name": "action to take", "input": "Action input"},
-    "Observation_1": "result of the action",
-    "Evaluation_1": {"evaluation": "your evaluation", "score": your evaluation score}
+    "Evaluation": {"evaluation": "your evaluation", "score": your evaluation score}
   }
   """
   
@@ -233,7 +239,7 @@ function progressValueEstimator(a::T1, state::T2)::Tuple{String, Integer} where
   the current situation and actions that can be three types:
   1) winestock[query], which you can use to find wine in your inventory.
   2) chatbox[text], which you can use to interact with the user.
-  3) finish[answer], which returns your wine reccommendation to the user.
+  3) reccommendbox[answer], which returns your wine reccommendation to the user.
 
   Given a question and a trajectory, evaluate its correctness and provide your reasoning and
   analysis in detail. Focus on the latest thought, action, and observation. Incomplete trajectories 
@@ -250,14 +256,17 @@ function progressValueEstimator(a::T1, state::T2)::Tuple{String, Integer} where
     "Thought_2": "But there is only 1 model that has the feature customer wanted.",
     "Thought_3": "I should check our inventory first to see if we have it.", 
     "Action_1": {"name": "inventory", "input": "Yiem model A"},
-    "Observation_1": "Yiem model A is in stock.",
-    "Evaluation_1": {"evaluation": "This trajectory is correct as it is reasonable to check an inventory for info provided in the question.
+    "Observation_1": "Yiem model A is in stock."
+  }
+  {
+    "Evaluation": {"evaluation": "This trajectory is correct as it is reasonable to check an inventory for info provided in the question.
                                     It is also better to have simple searches corresponding to a single entity, making this the best action.",
                       "score": 10}
   }
 
   Let's begin!:
   $(JSON3.write(state[:thoughtHistory]))
+  {Evaluation
   """
 
   # apply LLM specific instruct format
@@ -292,11 +301,16 @@ function progressValueEstimator(a::T1, state::T2)::Tuple{String, Integer} where
 
   _response = GeneralUtils.sendReceiveMqttMsg(outgoingMsg)
   _thoughtJsonStr = _response[:response][:text]
-  thoughtJsonStr = jsoncorrection(a, _thoughtJsonStr, responseformat)
+  expectedJsonExample =
+  """
+  Here is an expected JSON format:
+  {
+    "Evaluation": {"evaluation": "...", "score": ...}
+  }
+  """
+  thoughtJsonStr = jsoncorrection(a, _thoughtJsonStr, expectedJsonExample)
   thoughtDict = copy(JSON3.read(thoughtJsonStr)) 
-  latestEvaluationKey, _ = 
-    GeneralUtils.findHighestIndexKey(thoughtDict, "Evaluation")
-  evaluation = thoughtDict[latestEvaluationKey]
+  evaluation = thoughtDict[:Evaluation]
 
   return evaluation[:evaluation], evaluation[:score]
 end