Prithvi Datla

October 2026 to present. I take new ventures across robotics media, web3, real estate, and consumer products from zero to one. Teams of AI agents run the work. I decide and press send.

Outside KiloByte, August to October 2026.

• Jev vs LLMs, a model compared with other LLMs on extracting rule markers from RFPs. F1 0.64, against 0.57 and 0.52
• JevBag, built at Jevathon, the first Jev hackathon. A guard that catches dangerous commands, such as a force push or deleting a database, when an IDE runs on auto-approve, and blocks them. Work in progress, and it will be open-sourced
• Momito, a free, local dictation app for macOS
• The RAG Book, a field manual for AI engineers on retrieval, agents, and multi-agent systems. 292 free downloads, now paid on Amazon
• Signal, not slop, a free product thesis for a trust-based professional network, given away with no ask
• Jumper robot sim, a simulation of three KingKong Robotics Jumper robots, programmed entirely through natural language with the open-source Jumper SDK. October 2026
• Cheap Signal, when signals get cheap, the people who cared leave

Applied AI for business, January to August 2026. What we built for clients:

• Dinealog, a multi-language voice agent for restaurants that takes orders and manages reservations. In live betas across three DFW locations, answering missed calls alone brought in more than 37% extra revenue a day; now shut down
• A competitor price tracker for a specialty food store. It surfaced 30 to 70% room for price increases
• A multi-agent payroll assistant for an IT staffing company. It cut back office processing from 100 hours a month to 30, for every 100 employees and contractors on the books
• A menu generator for a scratch kitchen that caters daycares, the second largest in North Texas. It gave the operator 20 hours a month back
• A stock-out predictor for a coffee and cocktail business in Ohio. It ran at 90% accuracy and headed off about three stockouts a week
• ERP and CX platform POCs for a British automaker: gated, staged payments, spec locking, and end-to-end visibility into the build
• Typewriter Justice, a typewriter repair and sales shop in Keller, Texas

And what we built in the lab, for ourselves:

• CivicLLM, self-hosted and trained on a city's own docs; we built the demo using the City of Brownsville as the example
• GTM on GitHub, our whole outbound motion run out of a private repo
• BIOS, our internal second brain
• Momito, free offline dictation for macOS
• The $7 model, tuned on Tinker by Thinking Machines
• Buzz for GTM, the week we ran that motion on Block's Buzz platform

I ran GTM for both KiloByte and Dinealog. The team's first couple hundred cold emails and messages drew zero responses, so I stepped in: defined the buyers, then ran our shepherd strategy through restaurant association leaders, consultants, and food distributors who walked us straight to owners. LinkedIn response rates topped 60%, and we went from zero meetings a week to six to eight.

OUTCOMESSeven client POCs and six lab builds in eight months. A city government website rebuilt in a weekend. A $7 fine-tune that beat a frontier model on the client's own eval. Outbound rebuilt from zero responses to six to eight meetings a week.

I talked to customers, shipped, and built the team.

Chief Sales Officer, responsible for revenue.

• Rebuilt the sales motion: new collateral, AI-enabled selling, sales org cut from 6 to 2, pitch deck from 30 slides to 10
• Ran six campaigns across higher ed, online gaming, manufacturing, mid-market BFSI, IT services, and airlines
• Built the inbound engine: a podcast across LinkedIn, Spotify, and YouTube, plus webinars and a cost-saving estimator
• Ran partnerships and co-sell with dbt Labs, Oracle, and Snowflake, plus SIs and ISVs
• Prepared the data room for a Series A raise

OUTCOMESDemos and POCs booked up 30%+ quarter over quarter. The podcast closed the first customer at five-digit ARR. Ten months in, that customer acquired Heeddata as its services accelerator.

An experiment to crack distribution: food content, posted as Chutneybro.

OUTCOMESThe content went viral at 440K views.

Product, before and after the Wireless and FiOS merger.

• Post merger, Lead PM: helped ship a new multichannel sales platform that let wireless and wireline cross-sell for the first time
• The platform ran across retailers, big-box stores, contractors, sales reps, and authorized resellers
• Pre merger: owned the FiOS accessories store and its P&L. Managed sales, post-sales, and servicing; grew the store via partnerships, promotions, new product launches, an IoT focus, and merchandising updates

OUTCOMESThe cross-sell platform shipped as a $120M line of business. Managed the FiOS accessories P&L at $7M and grew it to $11M in 14 months.

MBA at Cornell Tech, New York City.

• Helped build and ship a traveler help and navigation app for JetBlue Airways at JFK Terminal 5: Bluetooth beacons, indoor mapping, time to gate, search for refreshments, and promotions for terminal stores. JetBlue later shipped these features in its customer app
• Pitched studio projects to over 20 VCs, including Techstars and USV

PM for the first two years, then Lead.

• Built and managed the wireless store catalog: every wireless device, accessory, and plan launch online from 2010 to 2016
• Helped build omnichannel from zero to one across sales, post-sales, and servicing
• Led a 13-member onsite-offshore team of data analysts and project managers
• Ran the daily release for the catalog team and rebuilt the workflow, release pipeline, and database behind it
• Ran the store through every iPhone launch, managing 70 to 80 million orders a day at peak

OUTCOMESHelped take wireless ecommerce revenue from seven figures in 2010 to about $4.8B by 2016. The daily release went from shipping 8 to 12 hours of work a day to 250 to 300. The catalog grew from about 15 to 20 SKUs to 4,500+. Throughput time cut about 60% over 24 months.

First product seat, a two-sided edtech marketplace.

• Revamped the website to Web 2.0
• Launched live support
• Ran a $50,000-a-month SEM budget

OUTCOMESTraffic from 4K to 140K monthly visitors. Revenue from $300K to $2.7M ARR in ten months. Conversion up about 47%.

Interim COO of a family wealth office, July 2024 to March 2025.

• Acquired a limo company, a fleet of 23 vehicles
• Acquired and renovated a $5M Manhattan penthouse

OUTCOMESFour storeys of renovation completed in 90 days.

Mexico's best Indian restaurant. Invested about $7,000 into a little blue house in Tulum and cooked myself.

• Built the menu from scratch and two outdoor kitchens by hand
• Built the website, ran the Instagram, cleaned, hosted, and served myself for four months before hiring staff and launching the menu out of beta
• Hired and trained an 8-person Mexican and Argentinian team

OUTCOMESWaste under 0.3% for a year. Revenue reached $800 to $1,000 a day. Closed with the perfect five-star rating intact (Google reviews).

Seven years as a host, then a hotel of my own.

AIRBNBEx-Airbnb Superhost, 18 quarters in a row.

• Seven properties across the US and Mexico, run solo until the first hire in year three

OUTCOMESA $9,000 investment turned into $1.3M in revenue over seven years. 2,700+ five-star reviews.

HOTELIERRaised $1.5M, bought land, formed a company, and started building our own hotel in Tulum.

• Navigated permits and regulation, hired 40 workers across trades, and ran the build as general contractor
• Exited at 30% completion; the project was acquired by a private equity firm and completed as Arka Tulum

OUTCOMES$1.5M raised and a hotel that opened as Arka Tulum.

First principles on retrieval, agents, evals, and failure modes - from a product and engineering lens. Written from my own builds.

AI explained from first principles, taught through one coffee shop, for readers who want to understand the machine, not just use it. 69 pages, free.

A LinkedIn newsletter and podcast. Lab notes from an operator.