Collin Kokotas

Staff Software Engineer

Elk Grove, CAcollinkokotas@gmail.comgithub.com/hoodiecollinlinkedin.com/in/collinkokotashoodiecollin.dev

Summary

Staff engineer and product-minded builder with 11+ years shipping web applications and AI products end-to-end — increasingly at principal scope. I conceived, pitched, and single-handedly built Velo™ — Recentive's flagship AI predictive-analytics platform — grew it into the company's #1 revenue source, and architected it into a platform multiple teams now build on. Deep in TypeScript/Next.js and agentic AI; at my best owning ambiguous, high-impact problems from zero to production, setting the technical direction others build behind, and lifting teams through architecture, documentation, and mentorship.

Skills

Languages: TypeScript / JavaScript (11+ yrs), Rust, Go, Python
AI & Agents: Vercel AI SDK, LangGraph, OpenRouter, Anthropic, OpenAI, Ollama, tool-calling, structured-output extraction, schema-aware query generation, fact-grounding, prompt engineering
Frameworks & Infra: Node.js, Next.js, React, Redux, GraphQL, SQL, Snowflake, PostgreSQL, Docker, AWS, GCP, Redis, CI/CD, Jest, Cypress
Practices & Tools: Monorepo architecture, type-safe codegen, CI/CD (GitHub Actions), Storybook, technical leadership & mentorship, cross-functional Agile/SDLC

Experience

Recentive Analytics — Staff Software Engineer

Sep 2024 – Jul 2026

Conceived, pitched, and built the proof-of-concept and v1 of Velo™, an AI-powered predictive-analytics and optimization platform for sports, media, and entertainment.
Grew Velo™ into the company's #1 revenue source — generating more than all other revenue streams combined — scaling to 12+ enterprise clients (several multinational) and a few hundred executive and decision-maker users.
Built the agentic-AI layer end-to-end (Vercel AI SDK, LangGraph, OpenRouter — Anthropic/OpenAI): tool-calling agents that generate and run warehouse queries via LLM-driven schema selection over a custom semantic-metadata layer I added to the Snowflake schema — an approach I adopted after vector RAG underperformed.
Designed a generic file-ingestion engine (docx/csv/pdf/xlsx, a few MB to multi-GB) that uses an LLM to derive each file's purpose and semantic metadata and materialize it into dynamic Snowflake tables the agents query.
Built “grounded facts,” a layer that mints verifiable facts from tool/query results and audits responses for unsupported figures — cutting responses containing ungrounded numbers from >60% to <10% across real customer threads.
Architected the platform (monorepo, build tooling, type-safe codegen) and authored the architecture docs, design specs, and diagrams that onboarded data science, data engineering, and application teams to build on it.

Mayan — Staff Software Engineer

May 2021 – Aug 2024

Technical lead for the core web app; wrote custom build tools generating strongly typed SDKs from backend services, cutting front-end effort and catching breaking changes before deploy.
Designed the Snowflake and PostgreSQL database schemas underpinning the product, and delivered third-party integrations across Amazon Ads / Seller-Central, GCS, Intercom, Looker, and Chargebee.
Built a Figma-driven React component library with a Storybook showcase, and added runtime metrics and error logging that measurably reduced error frequency and improved performance.
Established CI/CD (GitHub Actions) and a release process that eliminated broken deployments; guided QA to comprehensive Cypress end-to-end coverage, driving regressions per release to zero.
Contributed directly to other engineering teams' Python and TypeScript projects, unblocking cross-team delivery beyond my own product area.
As engineering & QA manager, led a daily cross-functional standup that cut support-ticket resolution from 1–2 weeks to 1–3 days, overhauled the technical hiring process, ran annual performance reviews, and coached struggling engineers back to standing through structured improvement plans.

AVB Marketing — Senior Engineer

Apr 2018 – Mar 2021

Front-end lead for an e-commerce platform powering 550+ websites; raised average Lighthouse performance from 12 to 92 and built a prerender service taking product-page SEO from 0 to 100.
Integrated Amazon Payments, Synchrony, and PayPal (+33% cart conversion); Dockerized the project, cutting new-developer onboarding time by 25%.

KeeperSecurity — Senior Engineer

Apr 2017 – Mar 2018

Rewrote the browser-extension autofill algorithm (~2s → <100ms) and cut site-compatibility failures from 20% to 3%.
Extended the autofill engine to support payment and address forms, broadening coverage across the sites users relied on.

VISA — Software Engineer

Apr 2015 – Mar 2017

Built a QA request-recording proxy that drove “cannot reproduce” tickets to near zero on the Express Checkout widget; optimized bundle size and client-side loading.

Selected Projects

ForgeDB

source2026

An application-database generator: write one declarative .forge schema and, at compile time, get a tailored Rust database, a typed TypeScript SDK, and a REST API with an OpenAPI 3.1 spec. It's a code generator, not an ORM or query engine — the output is specialized per schema over columnar storage, so there's no generic runtime to pay for.

typescript-to-rust (ttr)

2026

Language-level translator that compiles a strict TypeScript dialect into idiomatic Rust with true ownership semantics (borrows vs. owned values, &self / &mut self methods), with output verified by a real cargo toolchain rather than string matching.

Optigon

2026

Packages several interchangeable implementations of an operation (sorting, dictionary lookup, more to come) behind one interface, then learns per workload which is fastest via regret-scored adaptive dispatch. One Rust core, shipped as native addons to TypeScript (Node + Bun) and Python.

checked-rs

2024

Rust library that encodes arbitrary validation logic into the type system, with a proc-macro that generates specialized, self-validating integer types.

use-quantum-state