Ritesh
Pakala

Machine Learning Engineer / Applied AI Architect

Building Rao's personal AI apps for the Mac, Ambient, Craft, and Veil, and the AI systems underneath them: on-device models, retrieval, provenance, and safety. They credit the people whose work they draw on.

Rao Studios grew out of research I'd been carrying for years: NLP since 2016, on-device computer vision at Polarr from 2018, federated social since 2023, then Seer, Thread, and CLR across 2025 and 2026.

The work began in October 2025 with Seer, built outside working hours during my previous contract, and the apps followed in 2026. Rao Studios, my own company, became my full-time focus in October 2026, after that contract ended. The ideas, architecture, and core IP are my own; execution and completion are AI-assisted with Claude Code.

• Built Ambient (ambient.rao.nyc), a macOS reading companion that indexes what you read on screen into a local vector store and answers by voice or text, citing every answer back to its source, with transcription and speech running on-device
• Designed Rao Verified, a provenance record sealed by a Secure Enclave key that marks whether a person read, typed, or said something live, so only verified human input counts toward training and royalties
• Built Craft (craft.rao.nyc), an on-device coding agent for the terminal and the Mac that runs a 48B mixture-of-experts coding model (3B active) on MLX, with sub-agents sharing one set of resident weights under a measured GPU memory budget
• Shipped the Craft Agents store, where creators list agents from GitHub and keep 90% of sales (payouts through Stripe Connect, not yet open), on Rao's self-hosted Supabase API that also bills Ambient Plus
• Built Veil (veil.rao.nyc), which halts diffusion models mid-generation when they match a protected image: between denoising steps it compares tiles of the decoded latent against protected images and returns an attribution matrix, a royalty sheet, and a block verdict, with executors for FLUX.2 Klein, SD 1.5, and SDXL
• Authored Veil8, the open standard Veil's tiles are written in: a keyed, one-way encoding that turns each 32×32 tile into text a model can compare but no one can turn back into pixels, so protected images are matched without leaving the machine that holds them
• Building VeilGarde, Veil's detection and watermarking research: an AI-generated image detector that combines DINOv2 features with a camera-capture signature and retrains from stored features without keeping any image, plus a crop-robust watermark that adds nothing to the pixels
• Unified all three apps on one shared ML stack: a single on-device vector index with 1024-dimension MLX embeddings, a shared model home, and Rao Studios model mirrors on Hugging Face
• Designed each product site as a working in-browser mock of its macOS app, ported from the Swift sources with Claude Code's help, so visitors can use the app before installing it

Senior iOS Software Engineer

Contract iOS engineering role at Apple Inc. through INSPYR Solutions. Built native iOS tools and prototypes across Apple's platforms.

• Set up the initial tools and frameworks for building Sales training solutions
• Established frameworks and configured iOS applications for prototyping
• Provided iOS architecture guidance for AI-enabled prototypes
• Partnered with Machine Learning Engineering (MLE) and Design teams, translating design requirements into working prototypes

iOS Software Engineer, Platform Team

Contributed to architecture planning for the core mobile team's iOS platform, influencing multiple feature teams across the Walmart iOS app.

• Monitored and visualized app size metrics with Elasticsearch dashboards
• Investigated team-specific frameworks and the third-party dependencies integrated into the app
• Worked in an agile process with biweekly retrospectives, using Jira and Confluence

iOS Software Engineer

Worked on the 24FPS video app and its computer vision features.

• Rewrote the 24FPS video app from UIKit to SwiftUI, including Auth, Profile, and Video Filter Previews
• Led development of the Search/Explore feature for filter discovery
• Built animated transition previews and an infinite grid of before/after filter previews
• Achieved stable 4K video exports on iPhone 6s and later

iOS/ML Application Engineer

Photo editing software and enterprise computer vision models. Built mobile AI features and worked on-site with Samsung.

• Independently designed and built the Memoir app, using Core ML models for aesthetic ranking and auto-crop composition
• Worked on-site with Samsung at its headquarters in Suwon, South Korea, to integrate Machine Guided Composition and validate it on the Galaxy S10
• Jointly demonstrated <6ms inference with an SNPE model and a custom OpenCV implementation
• Achieved <500ms 2× upscaling (1080p → 2160p) on iPhone X
• Collaborated with research scientists on functional prototypes using Core ML models

Founder

Seed-funded menswear e-commerce startup. Built the storefront and led the first collection from design through production.

• Built the e-commerce storefront with Node.js, Express, EJS, and Firebase, hosted on AWS EC2
• Integrated Shopify for SKU management and payments
• Led design and production of a complete menswear collection from inception

Researcher

Used Python and SciKit to process high-resolution lunar surface images and identify optimal landing zones.

• Developed a novel method to identify optimal lunar landing zones from optical data alone
• Mapped crater locations and sizes without elevation data