YuvarajNagi
Building AI agents at Chewy and at home
I build AI agents that do real work, and the lab they live in.
Things I've built
The workshop
Drag the ring, or use the arrows. Click a card for the whole story.
For recruiters
At a glance
The 30-second version: where I am, what I've moved, and what I work with. The full story is in the journey below.
5+ yrs
in data, ML and AI
BI Engineer II
at Chewy, since Nov 2024
M.S. CS
Stevens · GPA 3.76
US
based
Impact
- ~0 minfrom a plain-English request to a production reportTails · Chewy
- 0%fewer loan defaults from my GBM premium modelAureus Analytics
- 0%better sentiment product for insurance clientsSentiMeter · Aureus
- 0co-ops and engineers I lead and mentorChewy
- ~0self-hosted services, with a watchdog that fixes itselfMy home lab
Case study · Chewy
Tails, in 10 minutes
The AI-powered metrics and reporting platform I designed and built for Chewy's analytics teams.
How a report gets made
Plain-English request
an analyst asks for a report
Agentic harness
context, memory, guardrails, loop, tools
Metrics catalog
one shared definition per KPI
Snowflake
the warehouse underneath
Production report
in about 10 minutes
Underneath: Airflow refreshes every KPI on schedule · runs on AWS ECS with S3 and ElastiCache
Problem
- A new report took analysts and engineers far longer than it should.
- Teams re-wrote the same KPI SQL in different places.
What I built
- Designed and built Tails end to end: backend, APIs and a central metrics catalog.
- The agentic harness Tails runs on: context, memory, guardrails, the loop and tool access.
- Airflow pipelines that keep KPIs fresh, on AWS (ECS, S3, ElastiCache).
Result
- Plain-English request to production-ready report in about 10 minutes.
- Teams onboard their own KPIs without duplicating SQL.
- Used across Chewy's analytics teams; I lead and mentor four co-ops and engineers on it.
The road here
Journey.
I started in computer science without loving it. Then I actually dug in, and I've been hooked on building things since.
By day I'm a Business Intelligence Engineer II at Chewy, where I built Tails: an AI-powered metrics platform whose agentic harness turns a plain-English request into a production report in about ten minutes.
By night I run a two-machine home lab that hosts my own AI agents: one makes weekly stock calls and gets graded on them, one answers my phone, one reads what I save and turns it into notes.
Off the keyboard: cricket, squash, table tennis, a good book, a long hike. I have a cat and a dog, and they have my heart.
Nov 2024 – Now
Business Intelligence Engineer II
Chewy · United States
- Designed and built Tails, an AI-powered metrics and reporting platform used across Chewy's analytics teams.
- Its agentic harness (context management that compresses tool results and summarises history near the limit; short-term memory in Redis and long-term in S3; a sandboxed, Tails-only scope as guardrails; loop engineering; and TAG, tool-augmented generation that retrieves data with tools instead of a vector database) turns a plain-English request into a production-ready report in about 10 minutes, work that used to take analysts and engineers far longer.
- Built the backend, APIs and central metrics catalog, so teams onboard their own KPIs without duplicating SQL.
- End-to-end Airflow pipelines for KPI refreshes; AWS (ECS, S3, ElastiCache) with Snowflake underneath.
- Lead and mentor four co-ops and engineers: roadmaps, design reviews, implementation calls.
Nights & weekends
Building AI agents
My home lab
- Running my own agents in production: stock calls, voice, notes, research.
- Everything ships through pull requests, auto-deploys and self-heals.
Jul 2023 – Mar 2024
Graduate Research Assistant
Stevens Institute of Technology
- Research with the finance department on automated market makers like Uniswap, toward a lower-fee currency trading platform.
- Co-wrote a 50+ page market study; built a Python + C++ mock AMM with a triangular fee structure.
- Added features and a cleaner UI to Stevens' high-frequency trading simulation system.
2022 – 2024
M.S. Computer Science
Stevens Institute of Technology · GPA 3.76
- Picked up the Bloomberg Market Concepts certification along the way.
Feb 2020 – Jul 2022
Data Analyst
Aureus Analytics · Mumbai
- Owned Household ID creation, the company's flagship product, end to end: scripts through monthly reports.
- Made SentiMeter, the sentiment-analysis product, 25% better for insurers.
- ML models for retention, cross-sell, fraud and household analytics: 20% efficiency gain; a GBM premium model cut defaults 20%.
- Daily analytics for Bharti AXA, Aditya Birla Sun Life Insurance and Landmark Insurance Brokers.
Oct 2019 – Jan 2020
Data Analytics Intern
Auxilo Finserve · Mumbai
- Built a website that streamlined loan onboarding and disbursement.
- Wrote a Python rule engine that flags high-risk loan applicants.
2015 – 2019
B.E. Computer Science
University of Mumbai
- Where I stopped tolerating code and started loving it.
Live agent
Ask me anything
It's an AI version of me that only knows the facts on this page. Watch how each answer gets made.
Under the hood
8/8 left today
- 1
Guardrails
8 questions a day · 400-char cap · per-IP rate limit
- 2
Grounding
16 facts from this page, ~1650 tokens, nothing else
- 3
Model
a small, cheap chat model
- 4
Answer
waiting for a question
system: answer as Yuvaraj, only from these facts
Name: Yuvaraj Nagi. Role: Building AI agents at Chewy and at home. Based in the US. Email: yuvaraj-nagi.jobs@gmail.com.
About: I started in computer science without loving it. Then I actually dug in, and I've been hooked on building things since. By day I'm a Business Intelligence Engineer II at Chewy, where I built Tails: an AI-powered metrics platform whose agentic harness turns a plain-English request into a production report in about ten minutes. By night I run a two-machine home lab that hosts my own AI agents: one makes weekly stock calls and gets graded on them, one answers my phone, one reads what I save and turns it into notes. Off the keyboard: cricket, squash, table tennis, a good book, a long hike. I have a cat and a dog, and they have my heart.
Skills: LLM agents, OpenAI Realtime, Prompt design & guardrails, Whisper, SQL / Snowflake, Airflow, Tableau, Pandas, R, ML models (GBM), Python, FastAPI, PostgreSQL / SQLite, Redis, Docker, AWS, TypeScript, React / Next.js, Tailwind, C++, Solidity.
Projects:
- Macro Insights (AI stock agent, 2026): Makes 5 stock calls a week for the US and India, then grades itself every Saturday. A formula scores ~4,000 stocks on fundamentals, momentum, earnings drift and liquidity. An LLM agent picks from the shortlist, sets entries, targets and stops, and writes a thesis. Every Saturday the week is graded against real closes and the lessons feed the next batch. A market-trend vote decides how many buys it is allowed to make. Stack: Python, FastAPI, React, SQLite, LLM agents.
- Strategy Arena (Backtesting lab, 2026): Pits trading strategies against 5+ years of data and promotes the winner into Macro. Rule strategies and Macro's own formula are replayed week by week with costs, a 70/30 held-out split and regime breakdowns. An evolution league breeds new weight sets; a sticky champion per market feeds Macro's Strategy Picks tab automatically. Stack: Python, Pandas, Parquet, D3.
- Call Claudy (Voice agent, 2026): I phone my coding agent, tell it what to build, and it calls me back when it's done. A realtime voice front end hands tasks to a coding agent on my server, keeps talking while it works, and announces results mid-call — hands free, no screens. Stack: OpenAI Realtime, WebRTC, Python.
- MindStash (Second brain, 2026): Save a reel, repo or article from your phone; get study notes and flashcards back. An iPhone share-sheet shortcut sends anything to MindStash. It transcribes videos, breaks down repos, explains the concepts and schedules flashcards. Stack: TypeScript, Bun, LLMs, Whisper.
- Recipe Box (Recipe importer, 2026): Turns recipe reels, TikToks and blogs into clean recipes, every line traced to its source. Pulls captions, transcripts and on-screen text from a link, assembles a structured recipe and shows where each ingredient and step came from. Stack: TypeScript, Next.js, LLMs.
One source of truth: the same file renders this page and grounds the agent, so it can't make things up about me.
Pick a question below or type your own.