Building things.
Breaking things.
Learning things.

Most people talk about AI. I'm the one arguing with it at 11pm because the pipeline broke. 4 years building things that work — and a lot that didn't, until they did. Not here to hype it. Just here to show you what actually happens when you build with it.

// |
Lalithya Rao Chavala — AI builder, pictured with a book in a bookstore
// professional overthinker, pictured mid-thought
“the pipeline broke · the fix didn’t”“why did this work yesterday”“built it from zero · now it runs itself”“the chatbot hallucinated · confidently”“ok this is the version we’re shipping”
“the pipeline broke · the fix didn’t”“why did this work yesterday”“built it from zero · now it runs itself”“the chatbot hallucinated · confidently”“ok this is the version we’re shipping”
About

I build AI systems for a living. Chatbots that don't gaslight users. Automation pipelines that actually save time. Content workflows running across 8 countries — because apparently that's a normal thing I do now.

Studied engineering, graduated with a silver medal, and somehow became someone who argues with LLMs for a living. My professors would be so confused.

Here's my hot take after building real AI stuff: 99% of AI content is either "it'll steal your job" or "it made me a millionaire." I'm the remaining 1% — the prompts that worked on the 6th attempt, the pipeline that broke on a Sunday, the time I automated something and then had to explain to my own laptop why it was wrong.

Basically if AI had a behind-the-scenes documentary, I'd be the person they interview in the break room. Not here to hype. Not here to scare you. Just building things, breaking things, learning things.

AI is a tool, not a personality structure first, then automate user empathy over metrics honest builder
4M+organic impressions8 markets, from zero
53,940+chatbot messagesMira, zero downtime
1,600+employeesConfluence intranet, built solo
500+ articles shipped 7 languages
8 international markets one person
3 live sites zero developers
The Work

Things I actually built

// built at MPOWER Financing · Mar 2024 – Apr 2026 · Innovation & AI Pod · Reports to CTO

GrowthAI Ops

Multi-Market Content Engine

8 international markets. 7 languages. Zero existing playbook, zero existing audience.

Built the whole system 0→1 — AI-powered content pipelines, an AEO/SEO framework structured to rank on Google and get cited by ChatGPT, Claude, and Perplexity, Claude localisation skills per market, and a templatised WordPress publishing flow that cut dev effort per article by 80%. Treated each market as its own bet: GSC + GA + Ahrefs decided what scaled and what died.

0organic impressions / 6 mo
0articles
0markets
0languages
What actually happened

The usual way would've taken years

Anyone who's done content at this scale knows the math: manual research, manual writing, manual publishing, per market — you don't hit these numbers that way, not this decade. The only way to grow exponentially across 8 markets was to stop treating content like a manual craft and build AI workflows that moved at the speed the growth required. The automation wasn't a shortcut. It was the only way the math worked.

AI Ops

Mira — Production AI Chatbot

A live AI chatbot answering real students' loan questions. Sole owner. No safety net.

Owned Mira end-to-end on a dual architecture — Yellow.ai conversational flows plus OpenAI GPT Assistants with RAG-based knowledge retrieval. Intent design, KB versioning with strict UAT→Production controls, monthly feedback triage, A/B tests on conversation flows, and the GPT-4.0→4.1 model upgrade: scoped, tested, validated, switched over with zero downtime.

0messages
0users
0uptime SLO
5m 50savg session
What actually happened

A chatbot is never actually "done"

Answers go stale. Flows break. Users find edge cases nobody planned for. Owning Mira meant being the person who caught it before it became a pattern — a KB update that quietly started surfacing the wrong loan terms, a flow that looped someone instead of resolving them. The job was never just building it once. It was staying close enough to catch it fast when it drifted.

Growth

Testimonials Page MVP

MPOWER.com had no testimonials page. No social proof, no trust signal, and SEO left on the table.

Spotted the gap in the credibility funnel, proposed the fix, then built and shipped it solo — sourced the data points and testimonials, researched the page structure, built the MVP with AI, raised the tickets, coordinated deployment, got it live. Ship → measure → iterate, the way agile is supposed to work.

~1 weekidea to live
0impressions / 6 weeks
Soloend to end
What actually happened

It was stuck in the queue. So I unstuck it.

This page didn't exist because the normal route — content team, then developers, then prod — moves at sprint pace, and this needed to exist now for SEO and trust. So I took it and built the whole thing myself with AI, start to finish, in about a week. I'm not claiming it's perfect. It didn't need to be. It needed to exist and start working — that's the agile bet: ship, learn, iterate. Don't wait for perfect to give you permission.

Growth

3 Live Websites, Zero Developers

International student search traffic worth capturing — and no engineering resource to capture it with.

Built three live satellite websites independently using AI tooling — borderlessloans.com, studyabroadloans.com, studyinusacolleges.com — full site architecture, SEO-structured content, navigation, categories. Ideation to live, no dev tickets, no waiting. Designed to catch student search traffic and funnel it home.

0live sites
0developers involved
100%AI-built
What actually happened

The AI got the bones right. The rest was me.

10web.io could generate a site skeleton fast — but the AI-generated content needed serious manual cleanup before any of it was actually publishable. "AI-built with zero devs" is true, but it doesn't mean zero work. It means the work shifted: from writing code to editing, structuring, and quality-controlling what the AI produced. That's the honest version of AI-native building — the AI does the first 70%, and the last 30% is where the judgment lives.

Systems

Company Intranet, From Scratch

1,600+ employees across the globe. No central knowledge home. Nobody had built one.

Designed and built MPOWER's entire Confluence intranet from zero — 13+ department spaces, onboarding guides, SOPs, governance frameworks, employee directory, the works. Sole ongoing owner: every update, every fix, every new space request comes through one person.

0employees served
0department spaces
1owner (hi)
What actually happened

Version one was complete. And nobody could find anything.

The first build looked finished — every space existed, every page was filed. And people still couldn't find what they needed, so they didn't use it. The fix wasn't more content. I redesigned the whole structure around how people actually search for things, not how departments sit on an org chart. Rebuilt it, and then it clicked — everything findable, everything used. Structure isn't what looks organised. It's what gets found.

SystemsAI Ops

WhatsApp Automation, All Channels

Repetitive outreach handled manually. Leads slipping through the cracks. No owner.

Took sole ownership of Wati (WhatsApp Business API) across the company — Marketing campaigns, Servicing (ACH, Collections), Refi, Payment Reminders. Webhook configuration, template management, campaign setup, performance dashboards. Five channels, one POC.

0channels owned
Solecompany POC
Zeroleads lost after launch
What actually happened

The team was drowning in messages a robot should send

Before this, the repetitive outreach — reminders, follow-ups, the stuff that absolutely should be automated — was being handled manually by the Customer Relations team. Which meant leads and follow-ups were quietly slipping through. I took ownership, automated it end-to-end, and once it was live we stopped losing leads across every channel — servicing, marketing, collections — without adding a single person. Automation isn't about replacing people. It's about letting them stop doing robot work.

Live from GitHub

Projects, auto-updating

// pulled live from github.com/lalithyaraochavala every time this page loads — no manual updates, ever

Play Zone

You've read enough. Play something.

Three small games, built into the page. Because a portfolio that says "I build interactive things" should probably contain some.

Whack the Bug

Production bugs are escaping. You have 20 seconds. Click them before they vanish — every one you miss ships to prod.

SQUASHED: 0 TIME: 20s

Ship It or Skip It

Twelve AI product ideas of questionable brilliance. Drag right to ship, left to skip. Your verdict at the end says more about you than about the ideas.

Trace the Line

This squiggle is my design signature. Hold and trace it from the dark dot to the coral one without wandering off. Steady hands win.

ACCURACY: —
Where it started
Morgan Stanley — Knowledge & IT Analyst
APR 2021 – DEC 2023 · VIA WILEY EDGE · BENGALURU
0KB articles migrated
0ticket volume reduction
0CSAT
0first contact resolution
0agents trained

Where I learned that knowledge systems live or die on whether people can actually find things — a lesson that came in useful roughly every day since.

Like what you see?
Let's talk.

lalithyaraochavala@gmail.com
Get in touch Download resume
Skills

What I actually work with

AI & LLM Systems

  • Claude — skills authoring, MCP, production pipelines
  • OpenAI GPT-4.1 — production chatbot, model upgrade end-to-end
  • RAG architecture — implemented and running in prod
  • Yellow.ai — intent design, flows, 8-market deployment
  • UAT-based LLM evaluation — real testing, not theory
  • Context engineering — agent memory, retrieval, constraints
  • Prompt engineering — daily, across multiple production systems
  • Midjourney / DALL-E — image generation for mockups & content

Automation & Building

  • n8n — hands-on orchestration, multi-step workflows
  • Wati / WhatsApp Business API — sole POC, 5 channels, webhooks
  • One-click LLM publishing pipeline — days to hours
  • Lovable / 10web — AI-generated UI prototyping, vibe coding
  • Figma — prototyping, wireframing, basic UI design
  • Agentic pipeline design — end-to-end workflow architecture

Growth & Analytics

  • SEO + AEO — ranked on Google AND cited by ChatGPT, Claude, Perplexity
  • Google Search Console — 8-market tracking, primary measurement tool
  • Ahrefs + SEMrush — keyword research, competitor analysis
  • Google Analytics + Data Studio — dashboards, funnel analysis
  • A/B testing — content formats and chatbot conversation flows
  • SQL — querying and data analysis
  • Loom — async video communication and documentation

Product & Ops

  • Confluence — intranet owner, 1,600+ users, 13+ spaces, built from scratch
  • Jira — epics, boards, cross-team delivery
  • Notion AI — user research synthesis, documentation
  • Agile delivery — sprints, standups, retrospectives
  • Stakeholder management — reported directly to CTO
  • SOPs & governance frameworks — built from scratch
Education

The paperwork

MBA — Human Resource Management

Andhra University, Visakhapatnam · 2024

First Class

B.Tech — Electronics & Computer Science

KL University, Vijayawada · 2021

Silver Medalist · GPA 9.49/10