I build systems that fix things. I bring structure to chaos. And drama.
The thread in every job, side project, and weekend build: find the signal nobody named, ship the system that handles it, then make sure it actually lands. Docs are easy. Adoption is the work.
Currently leading Product Operations and Customer Care Strategy at EarnIn (12M+ users). 12 years prior across Meta (Creator Monetization, ~8% of revenue), Apple, Deloitte (Strategy & Operations), Microsoft, and Ericsson. MBA, IIM Lucknow. CS, IIT Madras.
Selected outcomes across EarnIn, Meta, Apple, and Deloitte.
pzle
A daily 4×4 grouping puzzle with AI-themed words. Solving reveals the hidden theme behind the 16 terms and recommends one paper to read, turning the game into a daily AI literacy ritual. One Claude call per day, edge-cached for millions of plays at ~$1.50/month.
paper-radar
Daily arXiv tracker that scores ML papers by real-world traction. Tracks GitHub repos building on each paper, citation velocity, and time-decayed shipping signals, instead of just summarizing abstracts. Two-tier LLM pipeline: Haiku classifies 150 papers daily, Sonnet writes narratives for the top 10.
stackmap
A living architecture explorer for the AI ecosystem. Maps companies as blocks, organized into 12 layers, connected by typed wires, assembled dynamically around the solution someone is shipping. Stays alive via a five-agent pipeline (Scout, Classifier, Comparator, Synthesizer, Auditor), each with its own eval cohort.
digital-twin
An AI alter-ego platform. Upload your resume, docs, and links to build a private knowledge base, then chat with an agent that represents you. Includes PII detection, per-user RLS, and an MCP server so other agents can talk to your twin.
nybf
A companion chat product with multiple personas (Partner, Big Sibling, Friend). The interesting half is the autonomous eval loop: simulated user agents stress-test the app, a product-head agent triages findings, a dev agent writes code diffs, a QA agent re-simulates to validate, all in one command.
social-media-pipeline
Fully autonomous daily social-content pipeline. A scheduled n8n trigger kicks off a Mastra workflow that generates copy with Claude, synthesizes images and videos in Python, then publishes via the Publer API. Zero human touch per post. Runs on a private host, not a Next.js app.
Patterns in data, conversations, and systems that other people walk past. The unnamed thing is usually where the leverage is.
Not a deck. The actual pipeline, classifier, workflow, or metric framework that turns the insight into a behavior change.
Adoption is designed in, not bolted on at launch. Every team that has to live with the system actually uses it.
I run on a Rumsfeld matrix, known unknowns versus unknown unknowns. Three steps that survive every project:
Choreography, freestyle, and the occasional reel that goes harder than it should.
@narenkris →Training is currently paused, but the logbook is open. Single-engine, VFR.
Most weekend builds below get from idea to shipped in 1-2 sittings. The compression is real.
What I've read, what I'm flipping through now, and what's queued up next.