I’m cleaning a weather dataset in Google BigQuery where missing values in wind_speed and visibility were incorrectly stored as 0. I tried updating the table like this: UPDATE `my_project.weather_data.tracks_2023` SET wind_speed = CASE WHEN wind_speed = 0 THEN NULL END, visibility = CASE WHEN visibility = 0 THEN NULL END WHERE wind_speed =(Read More)
I’m working on an AI application that uses a large language model for question answering over internal documentation. While retrieval augmentation has improved factual accuracy, the model still occasionally generates confident but incorrect responses when the retrieved context is incomplete or ambiguous. A simplified version of the inference pipeline looks like this: retrieved_docs =(Read More)
I’m working on an AI application that uses a large language model for question answering over internal documentation. While retrieval augmentation has improved factual accuracy, the model still occasionally generates confident but incorrect responses when the retrieved context is incomplete or ambiguous.
A simplified version of the inference pipeline looks like this:
retrieved_docs = retriever.search(query, top_k=5)
prompt = f"""
Use ONLY the information below to answer the question.
Context:
{retrieved_docs}
Question:
{query}
"""
response = llm.generate(prompt)I’ve experimented with increasing retrieval depth, adjusting chunk sizes, and rewriting prompts, but there’s always a trade-off between factual accuracy, latency, and response quality.
For those building production AI systems:
I’m particularly interested in approaches that have worked well in production rather than benchmark experiments.
I’m maintaining a machine learning model that’s deployed in production, and over the past few months I’ve noticed a gradual decline in performance. The data distribution has shifted, but not enough to justify retraining the model every few days. Right now, I’m monitoring prediction confidence and basic evaluation metrics, but I’m unsure when retraining should(Read More)
I’ve been preparing for data analyst and data scientist interviews, and I’ve noticed that many interview experiences online focus on SQL, Python, and statistics. But I’ve heard that some companies ask open-ended business or case-study questions that are much harder than coding problems. For those who’ve been through multiple data interviews: What was the most(Read More)
I’m working on an advanced analytics project where the goal is to estimate the impact of a business intervention (pricing changes, targeted campaigns, and operational improvements). The challenge is that there’s no randomized control group, and historical data is affected by seasonality, market trends, and multiple simultaneous interventions. I’ve experimented with propensity score matching and(Read More)
I’m working on an advanced analytics project where the goal is to estimate the impact of a business intervention (pricing changes, targeted campaigns, and operational improvements). The challenge is that there’s no randomized control group, and historical data is affected by seasonality, market trends, and multiple simultaneous interventions.
I’ve experimented with propensity score matching and difference-in-differences, but I’m struggling to determine whether the estimated treatment effect is genuinely causal or simply capturing hidden confounding variables.
For those who’ve worked on large-scale causal inference problems:
I’m interested in hearing how experienced data scientists approach this in production environments, particularly when the results drive high-value business decisions.
