Mohit Khanduri · Product Analytics Engineer · India
I turn product and usage data into metrics teams trust, and I'm building toward AI on top of them.
I build the data layer behind product decisions: event and app data modelled in dbt, clear metrics for activation, retention and feature usage, and self-serve dashboards. Lately I'm extending it with AI, like letting teams ask their product data questions in plain English, checked with evals.
Model product data
I turn app events, users and accounts into clean dbt models, so teams can see activation, retention and feature usage without writing SQL.
Make metrics trustworthy
One agreed definition per metric, tests on every model, data quality rules and access control, so every team sees the same numbers.
Exploring: AI on top of analytics
Plain-English questions over the metrics layer, LLM tagging of user feedback, and evals that check the answers. Follow along in the lab.
How I build
The full approach →Define the metric before the dashboard
One agreed definition of an active user
A tracking plan before tracking code
Test every model
Self-serve beats a ticket queue
No AI answer ships without an eval
Live demo
Product analytics for a fleet-service SaaS app
Events tracked in PostHog, modelled in dbt, activation and retention dashboards, and an "ask the data" box with its eval pass rate shown. Follow the build log while it's being built.
- PostHog
- dbt
- DuckDB
- Metabase
- LLM + evals
Selected work
Product analytics for an AI SaaS product
Set up the data stack from scratch and gave the product team self-serve reporting on signups and usage of every AI feature.
Industrial energy companyTrusted metrics in a lakehouse
Owned lakehouse models with medallion layers, SCD Type 2 history and data quality rules, so business metrics stayed consistent.
Group of portfolio businessesOne set of metrics across several businesses
Consolidated ERP data from several companies into one warehouse, with shared metric definitions and four-tier row-level security.
AI lab and build log
Exploration, clearly labelled: small experiments in where analytics and AI meet.
AI research advisor
A LangGraph agent that asks clarifying questions, researches the web and recommends with sources. Every run is traced in Opik.
- LangGraph
- FastAPI
- React
- Opik
Governed data agent
An LLM agent over a medallion pipeline that goes through the same consent checks, data masking and filtered retrieval as any other caller.
- DuckDB
- Python
- litellm
Latest from the build log
- Scoped the flagship project. Usage analytics for a sample fleet and field-service SaaS app: PostHog events, dbt models, activation and retention dashboards, then an "ask the data" layer with an eval set.
About
I started in enterprise IT as a business analyst, moved into software and have spent the last seven years in data modelling and analytics engineering. The part I enjoy most is the moment a product team agrees on what a metric means and starts using it to decide. I've built analytics platforms at a SaaS startup, an industrial energy company and across a group of businesses. Now I'm focused on product analytics for SaaS companies serving small businesses, and I'm learning how AI can make that data easier to use.
Now
- Building the flagship: product analytics for a sample fleet-service SaaS app
- Learning PostHog event tracking, funnel and retention metrics, and A/B test analysis
- Exploring text-to-SQL over a metrics layer, and how to eval it
Contact
Open to product analytics and analytics engineering roles, India or remote.