Projects
Production-grade engineering systems built with zero fluff. Source code, architecture writeups, and design decisions are published here as each project ships.
Parallax
Blast Radius & Semantic Drift CI for dbt.
Parses SQL Abstract Syntax Trees (ASTs), maps downstream lineage across your entire dbt DAG in memory, and blocks the PR before a filter tightening or dropped column silently corrupts executive dashboards. Operates statically with no warehouse credentials, zero API latency, and no staging cloud bills.
AST Semantic Diff Engine
Leverages SQLGlot to parse changed SQL models into typed expression trees. Distinguishes cosmetic whitespace from breaking semantic drift: predicate tightening, condition loosening, column drops, and join structure mutations across 25+ SQL dialects.
In-Memory Lineage Traversal
Loads compiled dbt manifest.json into an in-memory NetworkX directed graph. Instantly computes full reachable downstream subgraphs (nx.descendants) reaching marts, reporting models, and executive BI exposures.
Multi-Hop Column Tracing
Performs multi-hop AST tracing through aliases and transformations. If an upstream staging model drops or renames a column, Parallax identifies every downstream model where schema contracts break before tests even execute.
8-Rule Deterministic Risk Engine
Assigns CRITICAL, HIGH, MEDIUM, or LOW severity using eight deterministic rules. Deliberately built with zero LLMs and zero network calls for 100% reproducible CI exit codes and automated pinned PR comment updates.
The "Innocent" Filter Tightening That Silently Corrupts Financials
SELECT order_id, customer_id, status, order_total * 0.95 as net_booked_amount, order_date
FROM raw_orders
WHERE status NOT IN ('returned', 'cancelled');SELECT order_id, customer_id, status, order_total * 0.95 as net_booked_amount
FROM raw_orders
WHERE status = 'delivered';