Added kalshi bet predictor example

8-application-demos/6-kalshi-bet-predictor/.gitignore

@@ -0,0 +1,4 @@
+.DS_Store
+.env
+.venv/
+.vscode/

8-application-demos/6-kalshi-bet-predictor/cerebrium.toml

@@ -0,0 +1,23 @@
+[cerebrium.deployment]
+name = "kalshi-bet-predictor"
+python_version = "3.11"
+docker_base_image_url = "debian:bookworm-slim"
+disable_auth = true
+include = ['./*', 'main.py', 'cerebrium.toml']
+exclude = ['.*']
+
+[cerebrium.dependencies.paths]
+pip = "requirements.txt"
+
+[cerebrium.hardware]
+cpu = 4
+memory = 36
+compute = "ADA_L40"
+
+[cerebrium.scaling]
+min_replicas = 1
+max_replicas = 2
+cooldown = 30
+replica_concurrency = 1
+scaling_metric = "concurrency_utilization"
+

8-application-demos/6-kalshi-bet-predictor/main.py

@@ -0,0 +1,170 @@
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer
+from typing import Tuple
+import requests
+import re
+from dotenv import load_dotenv
+import os
+from exa_py import Exa
+
+def getKalshiQuestion(market_ticker)->Tuple[str,str]:
+    url = f"https://api.elections.kalshi.com/trade-api/v2/markets/{market_ticker}"
+    try:
+        res = requests.get(url)
+        res.raise_for_status()
+        obj = res.json()
+        return obj['market']['rules_primary']
+    except requests.exceptions.RequestException as e:
+        raise RuntimeError(f"Error fetching Kalshi market data: {e}")
+
+def getKalshiOdds(market_ticker)->Tuple[str, str]:
+    url = f"https://api.elections.kalshi.com/trade-api/v2/markets/{market_ticker}"
+    try:
+        res = requests.get(url)
+        res.raise_for_status()
+        obj = res.json()
+        return obj['market']['yes_ask'], obj['market']['no_ask']
+    except requests.exceptions.RequestException as e:
+        raise RuntimeError(f"Error fetching Kalshi market data: {e}")
+
+
+class BetPredictor:
+    def __init__(self, model_name: str = "Qwen/Qwen3-4B-Instruct-2507"):
+
+        self.tokenizer = AutoTokenizer.from_pretrained(model_name)
+        self.model = AutoModelForCausalLM.from_pretrained(
+            model_name,
+            torch_dtype = torch.bfloat16,
+            device_map="auto"
+        )
+
+        load_dotenv()
+
+        self.exa = Exa(os.environ.get("EXA_API_KEY"))
+
+        print(f"Loaded model {model_name}!")
+
+    def _generate_response(self, prompt: str, max_new_tokens: int) -> str:
+        inputs = self.tokenizer(prompt, return_tensors="pt").to(self.model.device)
+        input_ids_len = inputs['input_ids'].shape[-1]
+
+        output_sequences = self.model.generate(
+            **inputs,
+            max_new_tokens=max_new_tokens,
+            pad_token_id=self.tokenizer.eos_token_id,
+            do_sample=False,
+        )
+
+        newly_generated_ids = output_sequences[0, input_ids_len:]
+
+        response = self.tokenizer.decode(newly_generated_ids, skip_special_tokens=True).strip()
+
+        print(f"Generated this response! {response}")
+        return response
+
+    def convert_rules_to_question(self, rules:str) -> str:
+        prompt = (
+            "You will receive a sentence that is a statement of the following type:"
+            "If <conditional>, then the market resolves to Yes"
+            "Convert the conditional to a yes/no question"
+            "Your response SHOULD ONLY BE a SINGLE line consisting of the yes/no question:\n"
+            "Do not add ANY preamble, conclusion, or extra text.\n\n"
+            f"STATEMENT: {rules}\n"
+        )
+
+        raw_response = self._generate_response(prompt, max_new_tokens=400)
+
+        return raw_response
+
+    def get_relevant_questions(self, question: str) -> list[str]:
+
+        prompt = (
+            "Based on the following question, generate a list of 5 relevant questions "
+            "that one could search online to gather more information. "
+            "These questions should yield information that would be helpful to answering "
+            "the following question in an objective manner.\n\n"
+            "Your response SHOULD ONLY BE the following lines, in this exact format:\n"
+            "1. <question 1>\n"
+            "2. <question 2>\n"
+            "3. <question 3>\n"
+            "4. <question 4>\n"
+            "5. <question 5>\n"
+            "Do not add ANY preamble, conclusion, or extra text.\n\n"
+            f"Question: \"{question}\"\n"
+        )
+
+        raw_response = self._generate_response(prompt, max_new_tokens=400)
+
+        relevant_questions = []
+        for line in raw_response.split('\n'):
+            line = line.strip()
+            if line and line[0].isdigit():
+                clean_question = line.split('.', 1)[-1].strip()
+                relevant_questions.append(clean_question)
+
+        return relevant_questions
+
+
+    def get_information(self, questions):
+        results = [self.exa.answer(q, text=True) for q in questions]
+        answers = [r.answer for r in results]
+        return answers
+
+    def get_binary_answer_with_percentage(self, information: str, question: str) -> Tuple[str, str, str]:
+        prompt = (
+            "Analyze the provided information below to answer the given binary question. "
+            "Based on the information, determine the probability that the answer is 'Yes' or 'No'.\n\n"
+            "--- Information ---\n"
+            f"{information}\n\n"
+            "--- Question ---\n"
+            f"{question}\n\n"
+            "IMPORTANT INSTRUCTIONS:\n"
+            "1. Your response MUST ONLY be a single line in THIS EXACT FORMAT:\n"
+            "   Yes: <YES PERCENTAGE>%, No: <NO PERCENTAGE>%, Explanation: <EXPLANATION>\n"
+            "2. Percentages must sum to 100%.\n"
+            "3. Do NOT include any preamble, summary, or additional text.\n"
+            "4. Provide a brief but clear explanation supporting your probabilities.\n\n"
+            "AGAIN, Your response MUST ONLY be a single line in THIS EXACT FORMAT: Yes: <YES PERCENTAGE>%, No: <NO PERCENTAGE>%, Explanation: <EXPLANATION>"
+        )
+
+        response = self._generate_response(prompt, max_new_tokens=800)
+
+        match = re.search(r"Yes: (.*?), No: (.*?), Explanation: (.*)", response, re.DOTALL)
+
+        if match:
+            yes, no, explanation = match.groups()
+            return yes.strip(), no.strip(), explanation.strip()
+        else:
+            raise ValueError(f"Failed to parse LLM response: {response}")
+
+    def predict(self, question):
+        relevant_questions = self.get_relevant_questions(question)
+        answers = self.get_information(relevant_questions)
+
+        information = ""
+        for i, v in enumerate(relevant_questions):
+            information += f"INFORMATION {i+1}: \n"
+            information += f"QUESTION {i+1}: {v}\n"
+            information += f"ANSWER {i+1}: {answers[i]} \n\n"
+
+        yes, no, explanation = self.get_binary_answer_with_percentage(information, question)
+        return yes, no, explanation
+
+
+predictor = BetPredictor()
+
+def predict(ticker: str):
+    rules = getKalshiQuestion(ticker)
+    question = predictor.convert_rules_to_question(rules)
+
+    predYes, predNo, explanation = predictor.predict(question)
+
+    realYes, realNo = getKalshiOdds(ticker)
+
+    if realYes < predYes: # undervalued
+        buyYes = True
+    if realNo < predNo: # undervalued
+        buyNo = True
+
+    return {"buy_yes":buyYes, "buy_no": buyNo, "yes": predYes, "no": predNo, "explanation": explanation}
+

8-application-demos/6-kalshi-bet-predictor/requirements.txt

@@ -0,0 +1,44 @@
+accelerate==1.10.1
+annotated-types==0.7.0
+anyio==4.11.0
+certifi==2025.10.5
+charset-normalizer==3.4.4
+distro==1.9.0
+exa-py==1.16.1
+filelock==3.20.0
+fsspec==2025.9.0
+greenlet==3.2.4
+h11==0.16.0
+hf-xet==1.1.10
+httpcore==1.0.9
+httpx==0.28.1
+huggingface-hub==0.35.3
+idna==3.11
+Jinja2==3.1.6
+jiter==0.11.1
+MarkupSafe==3.0.3
+mpmath==1.3.0
+networkx==3.5
+numpy==2.3.4
+openai==2.5.0
+packaging==25.0
+playwright==1.55.0
+psutil==7.1.0
+pydantic==2.12.3
+pydantic_core==2.41.4
+pyee==13.0.0
+PyPDF2==3.0.1
+python-dotenv==1.1.1
+PyYAML==6.0.3
+regex==2025.9.18
+requests==2.32.5
+safetensors==0.6.2
+sniffio==1.3.1
+sympy==1.14.0
+tokenizers==0.22.1
+torch==2.9.0
+tqdm==4.67.1
+transformers==4.57.1
+typing-inspection==0.4.2
+typing_extensions==4.15.0
+urllib3==2.5.0