-orange?style=for-the-badge){kind=icon}
I've spent over a decade in product management, most recently as Group Product Manager at InMobi on Glance AI Lock Screen. It's built into the phone, not downloaded: wake the phone and you get wallpapers, news and weather picked for you, without unlocking. It shipped on 250M+ phones across 8 markets. I built the product team there from zero to five product managers, and promoted two to Senior PM.
The products below are mine.
Three of these have a written case study beside them, where I set out the problem, the calls I made, what I gave up, and what I would do next. Mission Mode is itself the fourth. There's a fifth on Ship Gate, inside the nine tools below.
| Product | What it does, and what the case study covers |
|---|---|
| JobMagnet see it → |
Paste a job ad and your CV, and five AI agents (small programs, each with one job) working in a row build you a portfolio site styled for that company, then hand you the code to keep. About 96 seconds end to end. I built it in 7 days for the OpenAI × Outskill hackathon. The agents are open too, at jobmagnet-codex. The case study goes into the hard part. Recolouring one template just gives you eight versions of the same site, so the layout and the mood had to change too. The gallery is one CV across eight companies, and you can see whether it worked. |
| GrowFlowAI try it → |
It turns a messy meeting note into a list of tasks, and it works from Telegram, the web or email, so you can paste a transcript, forward a mail, send a voice note or photograph a whiteboard. I use it every day. In the case study I explain why I stopped trying to build a better task list. Tasks get dropped because typing one up costs effort at the exact moment you're busiest, so I let you forward the note from wherever you already are and the structure appears on its own. |
| Kahaani AI try it → |
It writes three publication-ready audio scripts in about 90 seconds for roughly ₹2, about two US cents, against the ₹2,000 and the two to three days a freelancer takes. I built it in four days. The case study covers the cost claim. I put it on screen with a plus or minus 15% band attached, because a buyer can act on a number they trust, and a round number with no band would have looked better and meant less. |
| Mission Mode see it → |
A product exercise I set myself on growth for people learning English in India, covering who to serve, what they're actually trying to do (jobs to be done, or JTBD), the growth model, and a prototype you can click through. This one is the case study. It's the artifact recruiters usually ask for, so I made it public. |
Glance AI Lock Screen: 250M+ phones, 8 markets, a product team I built from zero to five. Magicbricks Pay Rent: a ₹1 crore pilot to ₹50 crore a month at peak. A crore is 10 million rupees.
📄 Glance AI Lock Screen and Magicbricks Pay Rent, in full
Glance AI Lock Screen, InMobi (2022 to 2026, Group Product Manager from January 2023)
- US launch: 100K to 10M devices in 18 months, reaching 6M daily active users (DAU)
- Average revenue per user (ARPU) growing 25% year on year, and $10 to $15 across a user's lifetime (two different measures, which is why they don't multiply out)
- Day-60 retention 49% to 55%
- Phone makers (OEMs) and phone networks (telcos) we shipped on: Samsung, Motorola, Sharp, Xiaomi and Realme, plus Verizon in the US and SoftBank in Japan. Business development closes the deal and I led the product integration after it. The Motorola one became the pattern every later integration copied.
Magicbricks Pay Rent, Magicbricks (2015 to 2021, Product Manager)
The rent-payment product I ran at India's property marketplace.
- Gross merchandise value (GMV): a ₹1 crore pilot to ₹50 crore a month at peak. In its fastest stretch it doubled month on month.
- Payment success 95% to 99%, and settlement from about 48 hours down to minutes
- I owned the take rate, the cut we kept on each payment: 30 basis points, which is 0.3% of what went through, negotiated up from 0.25% as volume grew, with 0.35% in discussion when I left
- Three payment gateways integrated, with cashback wiring across five banks: HDFC, ICICI, Kotak, Standard Chartered and IDFC
I'm looking for a Group PM, Principal PM or Director of Product role, at an AI-first company, at a global company's India centre (a GCC), or at an Indian product-led company. The work I want is 0-to-1, which means taking something from nothing to a real product, along with the go-to-market (GTM) that follows. I'm based in Bengaluru and will relocate anywhere in India for the right role.
- 💼 LinkedIn, which is the fastest way to reach me
- 🌐 aksheywalia.in, my portfolio, with a voice AI you can talk to
- 📧 aksheyw [at] gmail [dot] com
I build with AI coding assistants, which are programs that run in a terminal, read the files in a project and edit them on instruction. Anthropic's Claude Code is my main one. I use OpenAI's Codex and Google's Antigravity alongside it, and they are good at different things: JobMagnet's five agents run on Codex, not on Claude. An agent here is a piece of software given one job and its own written instructions, not a person. I used to need a team and a quarter to ship a product. Now I need a weekend and an agent.
The catch is that a general-purpose assistant with no guardrails is where the bad work comes from, so I set the guardrails up once and they load into every project I open: standing rules, a bench of specialist reviewer agents I can call on, and small scripts that fire on the action itself. I've open-sourced that setup as nine repositories. They're mostly written instructions and small scripts rather than software products, which is what makes them portable.
| Tool | What it does |
|---|---|
| claude-code-ship-gate | Blocks a bad push. Before Claude can push to your main branch, it has to pass whichever checks apply to what changed, from the tests through to a code review, a security pass and a secret scan. It sits outside git, the standard tool for saving and sharing code, so the flag people normally reach for to skip a check has nothing to skip. Switching it off is a decision you make on purpose, in one of four documented ways. The case study covers where I stopped hardening it and why, and the separate mistake that shaped those four: I once switched it on for every repo at once. |
| claude-code-deep-review | Finds what one review misses. Keeps re-reviewing the same work from a different angle each time, and stops only when a whole pass finds nothing new. The first time I ran it on a release it took 28 rounds and found 14 bugs a single pass had missed, two of them bad enough to hold the release. |
| claude-code-pm-agents | Pressure-tests a big decision. 16 specialists I can call on, covering product requirements, growth, brand, app store optimisation (ASO), search engine optimisation (SEO), YouTube and communications. Nine of them form a council for a big call: eight answer it independently, never seeing each other's replies, and a ninth chairs without voting and prints the disagreement word for word instead of averaging it away. |
| context-bridge | Carries context between sessions. A small per-project wiki, just markdown files. At session close the assistant proposes what's worth keeping, you approve it item by item, and the next session picks up in about 30 seconds where the last one left off. |
| claude-code-guardrail-hooks | Catches four mistakes, and says what it misses. Four small scripts that fire on the action itself, not on the assistant remembering a rule, which is the part it drifts on. They cover a secret landing in a file, a style config loosened to pass a failing check, work carrying on after you said stop, and a helper agent's findings lost when the chat trims. The two that watch files only see the assistant's own edits, so a file written by a shell command gets past them. I say so in the repo, because a check you trust for more than it does is worse than no check. |
| claude-code-harness-backup | Survives a wiped laptop. Backs up your rules, agents, skills and settings, because reinstalling the app takes two minutes but the corrections you taught it are gone if you were the only copy. |
| claude-code-rules | Stops the same corrections repeating. The thirteen standing rules I load into every session, so I stop re-explaining myself every time it forgets. Be honest about what you haven't checked. Write the test first. Read the file before you ask me what's in it. |
| claude-code-learned-skills | Turns a solved problem into a shortcut. Twelve skills, all written after the fact, out of debugging sessions and testing, so the second time the same thing bites it costs minutes. Servers and deployment, machine learning (ML) pipelines, prompting, and a ten-category bug audit. |
| career-command-center-template | Makes each application easier than the last. The same idea applied to a job search: twelve skills that install in one step, so the tenth application starts from what the first nine taught you. |
AI & Agents
Build
These are the tools I work with, not one product's stack. The large language models (LLMs) behind my building are Anthropic's, OpenAI's and Google Gemini's; the live products themselves all run on OpenAI models. OpenClaw is an open-source personal assistant I run on my own server and message on Telegram. I use n8n to wire services together, and Vapi for voice. Underneath: React, with Next.js or Vite depending on the job, Supabase where a product needs a database or sign-in, Vercel for hosting, and GitHub Actions for continuous integration (CI), running the checks.
📋 Awards and certifications
Recognition
- InMobi Real Star Award, the company's highest individual contribution award, won twice (2022 and 2025)
- InMobi top-performer recognition, four years running (2022 to 2025)
- ISB Flagbearer Award (October 2021), top 1% of the MBA class
- ISB Alumni Association Board Member (2024 to present)
Certifications
- Claude Code: SE with GenAI Agents, Vanderbilt University (2026)
- AI Product Management, Duke University (January 2026)
- Agentic AI and AI Agents for Leaders, Vanderbilt University (January 2026)
- Generative AI Strategic Leader, Vanderbilt University (December 2025)
- AI Generalist Accelerator Program, Outskill (October 2025)
- Digital Product Management Specialization, UVA Darden (February 2021)
Last updated: 2026-08