Sai Alekhya Chandramouli

Column-level SQL lineage and impact analysis: given a column you want to change, it lists the downstream columns and dashboards that read it, resolving SELECT *, CTE chains, joins and UNIONs.

• In the demo, 10 models touch orders, but only 6 columns and 4 of 5 dashboards read orders.discount_pct
• Unresolvable stars are reported as unresolved rather than silently dropped
• 18 tests plus a browser demo of dashboard impact

• Python
• SQL
• JavaScript
• Node.js
• Mermaid
• GitHub Actions

An activity-based cost-to-serve model for a parcel carrier that runs the same cost pools under several allocation driver sets and flags customers whose profitability sign flips depending on the driver chosen.

• 12 of the top 50 accounts by revenue are contribution-negative under activity-based drivers; 19 change sign across driver sets
• Seven cost pools fully reconciled with zero leakage under every driver set
• Seeded fixture of 200 customers across three segments, covered by 17 tests

• Python
• Activity-Based Costing
• Sensitivity Analysis
• GitHub Actions

Markov removal-effect marketing attribution scored against simulated holdout experiments, publishing a per-channel bias factor that corrects daily attributed numbers and shrinks as evidence ages.

• Compares last-click, linear and Markov models against experimental truth across 60,000 simulated users
• Markov over-credits brand search by 2.34x versus true incrementality
• Seeded end to end, so repeated runs print identical output

• Python
• Markov Chains
• Attribution Modeling
• Experimentation
• GitHub Actions

Identity resolution where per-purpose consent gates the graph join itself, measuring match quality as precision against planted ground truth plus a household-confusion count.

• Resolves 1,432 records from 300 synthetic households
• Marketing purposes hold 1.0000 precision with zero household confusions
• Purpose-scoped traversal via a recursive SQL CTE

• TypeScript
• Node.js
• SQLite
• SQL
• GitHub Pages