Your business should run on a tenth of what it costs today.
I use AI to rebuild how an organization operates: product, growth, support, content, back office, engineering. The goal is not a cheaper tool here and there. It is the whole operating cost at one-tenth, and growth funded by what you stop spending. Not 20% better. An order of magnitude, as the minimum.
Message me on WhatsApp +91 97417 99611 — tell me what your business costs to run each month and I'll tell you where the tenth is.Same business. Same customers. One-tenth the cost.
Where the tenth hides
Most operating budgets are paying a premium on three things: intelligence and software you rent, growth spend that buys nothing, and people doing work that agents now do. Each one has an order-of-magnitude move available right now, and together they take the whole P&L down a decimal place.
Own the model
Open-weight models now match the API incumbents on most real workloads, including Indian-language ones. Benchmark them against your current provider on your own data, then run them on GPUs you control. Per-token cost drops by 10× or more; latency and data residency get better, not worse.
Spend on incrementality, not correlation
Growth budgets grow on dashboards that confuse “users who saw the ad also bought” with “the ad caused the buy.” Geo holdouts and proper attribution usually reveal that a large slice of spend buys nothing. Cutting it is the cheapest growth you will ever get.
Run the company with a tenth of the team
Agentic workflows mean a small, senior team runs what used to need departments: support, content, operations, finance, engineering. One codebase, one app, one set of agents doing the repetitive work, instead of parallel products, parallel processes, and parallel teams.
Done this at scale, not on slides
- Role
- Senior Vice President, Physics Wallah
- Scale
- One of India’s largest ed-tech platforms, millions of JEE and NEET students
- Scope
- Product, growth analytics, and technical infrastructure
At PW I led the move of a live voice AI tutoring product off paid APIs and onto a self-hosted stack: open-weight speech recognition, an open LLM, and open text-to-speech, running on GPUs we sized ourselves. Models were benchmarked on Hindi and Hinglish rubrics before switching, so quality went up while the cost structure changed completely. The product now sells an annual plan at the price competitors charge per month.
On the growth side, I replaced correlation-based spend decisions with geo incrementality tests, built hyper-local targeting down to the apartment-complex level, and fixed the instrumentation that everything else depended on. On infrastructure, I’ve worked on app architecture, search, and page performance for a site serving tens of millions of visits.
About
I started in semiconductor fabrication and VLSI design, where cost per unit is the only number that matters and “good enough” gets measured in defects per million. That discipline is what I bring to a company’s operating costs: measure everything, own what matters, and treat a 10× improvement as the baseline expectation rather than a moonshot.
I work as an operator, not an advisor. I’d rather spend a week inside your stack and your spreadsheets than deliver a deck.