LLM Predictive Analytics Integration
Budget: ₹12,500 – ₹37,500 INR
I already have a working large-language-model application and now I need it to do more than just generate or transform text. The next milestone is bringing true Predictive Analytics into the project so the system can forecast trends in our data and surface insights before they become problems.
Concretely, I want the LLM to understand a prompt such as “Show me the expected CPU, memory and disk utilisation for the next 24 hours” and then:
• craft the appropriate SQL query that pulls historic utilisation metrics (CPU, MEM, DISK) from my monitoring tables,
• run or hand that query off to the existing pipeline,
• feed the results into a prediction model (your choice of proven libraries like scikit-learn, TensorFlow, PyTorch, Prophet, etc.),
• return a concise, human-readable forecast plus the raw numbers so we can visualise them in our dashboard.
You will need to wire the ML inference step into the current codebase, expose it through the same API layer the LLM already uses, and document the new endpoint and model-training process. I am DB-agnostic for now—if your solution works with standard SQL it will drop straight in.
Deliverables will be:
• cleaned and well-commented code integrated into the existing repo,
• a reproducible training / retraining script,
• short README that explains config variables, dependencies and how to extend the prediction horizon,
• one brief screen-share or video walkthrough so the hand-off is smooth.
If this integration goes well there is more NLP and forecasting work on the horizon, but for now the goal is a reliable Predictive Analytics layer that my current LLM can call seamlessly.
Concretely, I want the LLM to understand a prompt such as “Show me the expected CPU, memory and disk utilisation for the next 24 hours” and then:
• craft the appropriate SQL query that pulls historic utilisation metrics (CPU, MEM, DISK) from my monitoring tables,
• run or hand that query off to the existing pipeline,
• feed the results into a prediction model (your choice of proven libraries like scikit-learn, TensorFlow, PyTorch, Prophet, etc.),
• return a concise, human-readable forecast plus the raw numbers so we can visualise them in our dashboard.
You will need to wire the ML inference step into the current codebase, expose it through the same API layer the LLM already uses, and document the new endpoint and model-training process. I am DB-agnostic for now—if your solution works with standard SQL it will drop straight in.
Deliverables will be:
• cleaned and well-commented code integrated into the existing repo,
• a reproducible training / retraining script,
• short README that explains config variables, dependencies and how to extend the prediction horizon,
• one brief screen-share or video walkthrough so the hand-off is smooth.
If this integration goes well there is more NLP and forecasting work on the horizon, but for now the goal is a reliable Predictive Analytics layer that my current LLM can call seamlessly.