“I have no special talents. I am only passionately curious.”
— Albert Einstein
Aaron Kankipati
I'm a Product guy by profession — seven years owning digital banking products end to end. I build autonomous AI systems on the side, because the only way to truly understand a tool is to ship something real with it.
Mobile Apps
Two Android apps, designed and shipped solo — one live in production, one in closed testing ahead of launch.
Moodfilm
In ProductionAaron Kankipati
AI film curator — tell it your mood, get a film matched to the moment.
Trap Tap
Closed TestingAaron Kankipati
A premium arcade guessing game — dodge hidden traps, fill the board, collect relics.
The Laboratory
By day I'm a Product Owner at Wells Fargo. These are built on late nights and weekends — the same product discipline applied to things I wanted to exist, each taken all the way to shipping.
Moodfilm
I am not a developer. I built this anyway. Product thinking over code.
An AI film curator that starts with one question — how are you feeling right now? Tell it your mood, pick a language, and it finds a film matched to that emotional moment.
Try Moodfilm v2 →How are you feeling
today?
Claude · film & streaming data · serverless edge
I used my own product.
Not from a code review — from opening the app the way a real user would.
Every poster showed the wrong film.
Wrong image on every card — every pick felt unreliable before you read a word.
Same films for every mood.
The whole premise is emotional relevance. Same blockbusters for every mood, and there's no product.
No Hindi or Telugu support.
Missing in V1, it became the #1 request after launch. Users set the roadmap.
I hit a ceiling.
So I redesigned around it.
A second film-data source gave richer results — but two libraries pushed 170 calls per session against a 50-call free-tier limit. The easy move was to pay. Instead I scoped it down. Batched and parallelised everything down to 22 calls max. Same data, still free.
Poster now always matches the film
Claude returns the exact film, so the right poster loads every time. No guessing. Wrong poster eliminated.
Language selection from user feedback
Any / English / Hindi / Telugu, applied at every step. Top request delivered.
Recommendation engine rebuilt
A psychological mood model with 900k+ films now reachable. Core promise restored.
“Did you know?” loading cards
30–60s of loading is enough to lose a user — so it holds 20 rotating film facts.
Streaming availability on every card
Region-aware, with a direct link. The job isn't done until users know where to watch.
One tap to share your pick
Native on mobile, clipboard on desktop. Organic reach built in, not bolted on.
“Tools change.
Product thinking does not.”
Understand the cost before optimising it. Moodfilm makes two Claude calls per session — one to map mood to a cinematic profile, one to select films and write the explanations. In V1 both ran on Sonnet with output ceilings set generously and never actually reached.
Match the model to the task. Mood analysis is pattern recognition and JSON extraction, not deep reasoning — so call one moved to Haiku, ceiling cut 900 → 400 tokens. Call two stayed on Sonnet where reasoning matters, but its ceiling dropped 2,500 → 1,400, since six explanations at ~200 tokens is 1,200. The rest was headroom that could never be used.
Cut the input, not the quality. The candidate pool dropped from 10 films per genre to 6, and overview text from 120 to 80 characters. Selection quality didn't change — the input size did. These look like engineering choices; the difference is knowing why each number exists.
The result. 54% cheaper per session, about $1.80/month at the current pace. The data sources and hosting all sit on free tiers — so the only bill is the Claude API, now a deliberate number, not a default.
Neon Bull
I gave an AI ₹10,000 and full control to trade the Indian stock market.
No human approves any trade. Claude reads real NSE data twice a day and makes every call itself. It's the proving ground for Project Zeon — prove the engine on paper before it ever touches real capital.
The Architecture
It runs entirely on its own. No server. No manual trigger. NSE holidays auto-detected and skipped.
Yahoo Finance — free, real NSE prices, no API key needed.
Claude Sonnet — RSI, SMA & volume analysis per session.
Cloudflare KV — portfolio, trades & P&L history.
Cloudflare Pages — dashboard with live experiment data.
Built Honestly — Not Financial Advice
- ✓Groww brokerage charges simulated on every trade — realistic P&L including STT, DP charges & GST.
- ✓Auto-pauses if portfolio value drops below starting capital.
- ✓Stock names fully hidden from the public — no SEBI compliance issues.
- →Cloudflare Pages + Workers + KV — zero infrastructure cost.
A controlled version of a bigger idea. Neon Bull is the proving ground for Project Zeon — a live-brokerage system I'd already wired up but deliberately paused. Same execution logic, real market, real prices, real brokerage. The only thing not real is the capital. Prove the engine before it touches real money.
The autonomy constraint forced every decision. Every trade is entirely autonomous — no approval, no veto, no second-guessing. The moment I can intervene, the experiment answers a different question. So I designed around never intervening.
The math had to work before the model did. V1 was structurally unprofitable — at ₹10,000 capital, a 1% move on one share earned ₹14 while a round trip cost ₹45 in brokerage. The fix was a product decision about the instrument universe and input size, not a smarter prompt.
Cost engineered down 99%. Across four versions I drove running cost from ₹41/day to ₹0.40/day by trimming inputs and matching effort to the task — without dropping decision quality. V1 trades twice daily; V2 moves to four sessions with a refined signal model.
Trap Tap
Tap the safe button. Dodge the hidden trap. Fill the board.
A premium React Native arcade game. Each round you tap one of two numbered buttons — one is safe, one is a hidden trap. Clear the board to beat a level, across 30 of rising difficulty.
Trap Tap
The Guessing Game
Adventure — one-life runs that drop rare relic “Tazos.” Classic — relaxed play with saved progress.
Every Tazo is a unique, seeded SVG disc across four rarity tiers — Bronze to Platinum. No two alike.
Momentum tiers, multipliers and reward feedback that make a hot streak feel earned.
Adapts to player behaviour, so the board can't be gamed by tapping the same side every round.
Currently in closed testing, ahead of launch.
Built with AdMob monetization, share-card generation and persistent collection storage.
About Me
Aaron Kankipati
Senior Product Owner · Team Lead · Digital Banking
Hyderabad, India
Product Owner with 7+ years across digital banking and fintech — most of it spent owning a single product end to end, from discovery and user stories to backlog, go-live and the accountability that comes after. I coach teams, build automated AI systems for fun, and stay relentlessly curious about how tomorrow's tools fit today's systems.
Senior Product Owner / Team Lead
Sr. AVP Nov 2025 — PresentWells Fargo · Hyderabad
Leading the Certificate of Deposits core-banking migration across two cross-geo Scrum teams (India + US). Own the backlog and discovery; accountable for change governance, rollback planning and release readiness on a critical legacy decommission.
Digital Product Manager / PO
AVP Jan 2024 — Nov 2025Wells Fargo · Hyderabad
Owned the Small Business Digital Deposit Application end to end. Cut onboarding from 18 to 9 minutes, drove a 66.5% fully-digital completion rate, and unified lending + deposit systems. Spotlight Award.
Business Analyst / Feature Lead
Manager Aug 2022 — Jan 2024IDFC First Bank · Hyderabad
Owned the conversational-AI suite — chatbots, IVR and a WhatsApp bot that hit a 60% lead-conversion rate and 30% loan disbursement. Wrote specs, ran UAT, and resolved incidents across engineering and ops.
Product Owner / Deputy Manager
First PO role Jan 2020 — Jul 2022Piramal Finance · Hyderabad
Owned the digital home-loan platform on Salesforce. Reduced KYC verification to under 3 minutes via OCR automation and built the onboarding framework for the sales team. Disha Champions of Change Award.
Sales Management Trainee
Jun 2019 — Jan 2020Piramal Finance · Hyderabad
Front-line work with home-loan customers — the first-hand understanding of customer pain that shaped every product decision I made later.
Exec MDP, Product Management
IIM Lucknow · 2020–21
PGDM
Institute of Public Enterprise
B.Com.
Little Flower Degree College
- AI for Product Management · Pendo
- Business Analyst Foundation · Kore.ai
- Design Thinking Practitioner · Wiley