Dhiraj
Chaudhary

Multi-agent system (LangGraph) that diagnoses job failures and drafts the fix — opens a GitHub PR for code issues, a one-click workflow for infra/data. Grounded in per-job profiles; human-in-the-loop by design.

The precursor: an LLM engine for data-file incidents. TPAs send hand-made Excel/CSV that breaks the pipeline — a renamed column, a changed date format — and it diffs expected-vs-actual, proposes a fix-the-file-or-fix-the-code action a human approves, then ingests instantly.

Two-agent data-quality system: one inspects a table, infers its domain, and proposes 10–100 rules (with tools to fetch stats, sample rows, and test-execute candidates); a second runs daily and reasons over rule output + exploratory stats to give a plain-English health verdict.

LLM system that auto-generates the 15–20 pipeline modules each new insurance deal needs — built on a canonical medallion model and grounded in prior deals' code. Shipped before Cursor/Copilot existed.

LLM pipelines that migrated 2,000+ Sybase procs → Redshift, 500 SAS jobs → Spark, and 1,000 Tableau dashboards via guided XML rewrites + a visual-diff system. Human-validated throughout.

Secure multi-language PDF/PPTX translation built from parts — parses every text + layout element, translates through open-source models running on-prem (nothing leaves the VPC), then reassembles into an identical-looking document. Verified by a round-trip eval loop scoring semantic similarity + sentiment divergence.

Config-driven Spark/Scala ETL framework I scaled into KKR's most-used internal utility. Self-service job generation, a validator producing 5M+ data-point checks, complex features behind single config flags.

Real-time ingestion for 300–400k tiny JSON files per load — jobs that once ran 17–18 hrs/day. Two wins: a Spark-distributed file move (5 hrs → 2 min) and RDD-level normalization that fixed list-vs-dict key drift before the DataFrame, after Lambda + pandas kept running out of memory.

One of the lead architects of KKR's move off 24/7 EC2 onto Glue / Lambda / EMR — designed the architecture, proved feasibility, built the v0, then enabled the org to migrate. Custom tooling where out-of-box couldn't: S3 trigger handler, dependency watcher, operational job handler.

Self-service platform to migrate any source DB — Oracle, Snowflake, DB2, Redshift — into Apache Iceberg, creating a unified, cheap data fabric. JSON-driven, built on the Spark framework.

Solo full-stack platform to track and sign off data variances across multiple databases — Flask end to end, with automated email notifications and approval routing. Shipped to AWS after passing the enterprise Architecture Review Board.

Keyboard-first 'mission control' for running many AI agents at once — drag-drop task lanes per agent, a Cmd-K command palette, and PM-grade timeline / radar / kanban views. Built it to manage my own parallel Claude runs.

Native iOS hydration tracker (SwiftUI) — animated progress styles, 90-day history, smart reminders, streaks, and full VoiceOver/dark-mode support. Built App-Store-ready, MVVM + OSLog.

Serverless student social platform — FastAPI on AWS Lambda with Cognito auth, API Gateway, and S3, plus a Vite/React front end. Multi-stage dev/staging/prod infra via the Serverless Framework.

A Python package (published to PyPI) for statistical operations and plots over binomial & gaussian distributions, built with OOP on pandas / NumPy / matplotlib.

Undergrad research: a Flood-It solver algorithm in Python (presented at the Mathematics Association of America) and a bio-robotics study on hybrid artificial/biological structures (SJC research symposium).