AI Data Analyst
Business questions answered in plain English, with a workflow built around 25 agents and 45 skills.
View projectSignal / noise
Emerging cluster
potential segment
potential segment
Model trend
signal
signal
Higher variance
noise
Response (standardized)
Feature 1 (standardized)
noise
AI Data AnalystBusiness questions → actionable answers
From data to decisions
Open to relocation & remote opportunities
Let’s find
the next signal.
↗Selected projects.
One analytics system. One hands-on data lab.
Explore the thinking and try the work.
Interactive Data Lab
Explore regression noise and classification thresholds in two working experiments using synthetic data.
Open the labExperience
Working with
the real world.
From financial transactions to everyday reporting.
Reliable data is where good decisions begin.
Less manual work.
More time for analysis.
Automated data entry, cleaning, and validation using Python and Pandas. Refactored reporting queries with indexing and CTEs, and built an Excel data-quality dashboard with Power Query.
~40%less manual processing time
~30%faster SQL execution
Approximate improvements reported in my résumé.
A closer look at the work
- Automation
- Python and Pandas workflows for data entry, cleaning, and validation.
- Query performance
- Indexing, common table expressions, and SQL refactoring for reporting queries.
- Data quality
- An Excel dashboard using Power Query and pivot tables for automated checks and weekly reporting.
A closer look at
financial risk.
Validated financial transactions with Python and SQL for compliance reporting. Supported ETL, analyzed spending patterns, and tuned queries used in fraud reviews.
500k+financial transactions validated
10+high-risk anomalies identified
A closer look at the work
- Validation
- Used Pandas, NumPy, and SQL to improve transaction records for compliance reporting.
- Fraud reviews
- Tuned SQL queries to surface high-risk anomalies for the security team.
- Pattern analysis
- Supported transaction-data ETL and used descriptive statistics to examine spending patterns.
From question
to a clear answer.
A closer look at the AI Data Analyst project.
Question → evidence → answer
Illustrative workflow25 agents / 45 skills
Agentic analytics
AI Data Analyst
An agentic analytics system that connects natural-language questions to data warehouses and explains the findings in plain English.
Ad-hoc analysis: from hours to minutes.
More code, experiments, and work in progress.Visit GitHub
Interactive data lab
Don’t just read it.
Explore it.
Small experiments in how models behave.
Move a control. Watch the trade-off.
Linear regression120 sample observations
Observations Fitted line
Experiment with uncertainty
How much signal
survives the noise?
The underlying relationship stays the same. Adding noise makes individual observations less predictable, even when a trend remains.
Clear signalMore uncertainty
R² · in-sample fit—
R² measures how much variation the fitted line explains in this sample. It does not measure performance on unseen data.
Decision threshold200 labeled sample records
Positive class Negative class
Experiment with a decision
Every threshold
has a trade-off.
Records at or above the threshold are flagged. A lower threshold finds more positives, but can also flag more negative records.
Flag moreFlag fewer
Precision—
Recall—
Precision: how many flagged records are positive. Recall: how many of all positive records are found.
Educational demos using deterministic synthetic data. These are not results from employer systems or the AI Data Analyst project.
My approach
A better question.
A clearer decision.
How I approach an analytics problem,
from the first question to the final explanation.
- 01A focused question
Define the decision
Start with the business question. Clarify who needs the answer, which metric matters, and what would change as a result.
- 02A reliable foundation
Understand the data
Examine quality, missing values, and possible bias. Clean and validate before treating a pattern as evidence.
- 03Evidence with context
Test the explanation
Explore patterns, choose a useful baseline, and evaluate a model or hypothesis with the right measures.
- 04An informed next step
Make the result usable
Explain the finding, its limitations, and the next action. Use a dashboard, a clear chart, or an automated workflow where it helps.
Toolkit
The right tool.
For the right question.
Analysis & engineering
Python, Pandas, SQL, PySpark, R, NumPy, dbt, Snowflake
Machine learning
Scikit-learn, Statsmodels, XGBoost, regression, classification, clustering, feature engineering
Business intelligence
Power BI, Tableau, Looker, Excel, Power Query, dashboard design
Experimentation & cloud
Hypothesis testing, power analysis, cohort and funnel analysis, AWS, Azure, Databricks
Explore the full technical stack
- Python & modeling
- Pandas, NumPy, Scikit-learn, Statsmodels, logistic regression, random forest, XGBoost, gradient boosting, K-Means, hierarchical clustering
- Product & risk analytics
- Cohort and funnel analysis, retention, user segmentation, conversion optimization, behavioral risk signals, precision, recall, AUC-ROC
- Statistics & experiments
- A/B testing, hypothesis testing, multivariate testing, power analysis
- Cloud & data platforms
- AWS S3, Glue, Athena, SageMaker, Azure, Databricks, Snowflake, dbt, PySpark
Behind the analysis
Hi, I’m Harshit.
I’m a data analyst based in India, working across analytics, automation, and predictive modeling. I enjoy making complex data easier to understand—and more useful to the people making decisions.
My experience spans banking and analytics, with a focus on data quality, efficient SQL, and Python workflows. I’m open to relocation and remote opportunities.
Download my résuméEducation
Bachelor of Computer Applications
IMS Noida · 2019–2022
Database management, statistics, and data structuresCertifications
- Microsoft Power BI Data Analyst Associate PL-300
- Google Data Analytics Professional Certificate · Coursera
- Python, C & C++ IIT Bombay