Shipping it
solo since 2024
Agentic AI Builder · Data-Driven Insights · Currently @ Cuemath
$ whoami
Gagandeep Singh — agentic AI builder · data-driven insights
$ claude status
▸ agents building · gates green · main protected
▸ nothing ships on vibes — verified in the running app
$ current
AI Builder @ Cuemath — live production platform, end to end
$
You're inside the screen now — keep scrolling ↓
AI is the engine.
Process is the moat.
I stopped writing code and started writing constraints. The agents build; I decide — what ships, when it's ready, and where the data says to aim next. Currently an AI Builder at Cuemath, running a live production platform end to end.
Where I add value
Agentic AI engineering
I orchestrate teams of AI agents under sprint plans and review gates — up to ~20 agents with a master orchestrator, explicit file ownership, and adversarial review. One person shipping at the pace of a small team. This is the craft I'm building a career on.
Data-driven insights
I model funnels as state machines and run drop-off RCA, Pareto, and source-attribution analysis — analytics that shape strategy, not just report on it. Every analysis lands as a decision brief: finding, recommendation, counterfactual.
Production ownership, proven
Currently as AI Builder at Cuemath: I design, ship, and operate a production platform serving live users end to end — role-specific experiences, multi-stage journeys, scheduling, engagement automation. Not a prototype: live, in production, mine to run.
Discipline that scales
One playbook for everything I ship, professional or personal: nothing merges on vibes. Four hard gates before anything is PR-ready, verification in the running app, and a hook that keeps even the AI off main.
Dangerously good shipping
Operations Platform
Production platform behind internal operations at Cuemath. Full-stack Next.js + Postgres serving live users end to end, plus the analytics layer that shapes strategy. Shipped via multi-agent Claude Code sessions with structured review gates.
Spend Tracker
Personal finance dashboard with Telegram bot input, calendar heatmap, savings goals, and a weekly Claude Haiku digest. Bento layout, single-accent palette, GitHub Actions CI + branch protection.
Kortex
Next.js app that ingests PDFs and generates spaced-repetition flashcards with Google Gemini — SM-2 scheduling, learn/practice flows, streaks, and a social feed. Neon Postgres, serverless on Netlify.
Self-Healing ML Pipeline
Novel ML pipeline that autonomously detects model failures, performs root-cause analysis, and applies adaptive recovery strategies. Companion VS Code extension for real-time monitoring. Documented across five phases culminating in an IEEE-format paper.
Every project here — including this site — was built with Claude Code.
The full method — guardrails, pipeline, and the thinking behind it — lives in the studio.