Initializing network

Arvini Consulting · AI Consulting Studio

Ship fast, reliable,
trustworthy LLM products.

Ship fast, reliable, trustworthy LLM products. I'm Arvin Ansari — currently a senior AI software engineer, helping teams at every stage bring context-aware, evaluated, guarded LLM systems to production.

Start a project

Valencia, Spain · Available for consulting, worldwide (remote)

Scroll
About

Senior AI engineer, trusted partner on hard AI problems.

I'm Arvin — a senior AI software engineer with 7+ years building software and a growing focus on making large language models production-safe and efficient.

Through Arvini Co I help startups and product teams turn LLM prototypes into dependable systems: composing context that actually fits, evaluating what matters, and putting guardrails around the unpredictable.

Portrait of Arvin Ansari seated outdoors in a blue striped shirt, warm sunlight and a green forest backdrop.

7+

years engineering

3+

years in LLM/AI

90%+

cost/memory reduction

500+

connections

What I can help with

Ways to work together

01

LLM Product Advisory

Architecture reviews, context-window strategy and model/tooling selection for teams building on LLMs.

02

Evaluation & Trust

Design evaluation harnesses, guardrails, sandboxing and red-teaming so you can ship with confidence.

03

Performance & Cost

Token/caching/inference optimization that cuts spend and latency while keeping quality high.

04

Full-Stack Delivery

Seven years shipping production React + Node systems, end to end — from monorepo to deploy.

Focus

Where I go deep.

Two areas where engineering judgment separates a demo from a dependable system — and where I spend most of my consulting time.

01specialty

LLM Context Engineering

Context is the product. Make it precise and cheap.

  • Token & context-window budgeting — keep every prompt on-model and on-budget.
  • RAG + retrieval optimization: chunking, entity-linked graphs, reranking.
  • Caching, streaming & inference tuning for latency and cost.
  • Long-context memory design for agentic multi-turn workflows.
02specialty

AI-Trust & Evaluation

Ship AI you can actually defend.

  • Evaluation harnesses: benchmarks, rubrics, regression suites.
  • Guardrails, sandboxing & tool-policy enforcement.
  • Alignment & safety patterns: red-teaming, audit trails, evals-as-tests.
  • Observability for model behaviour in production.
Experience

Seven years shipping production systems.

From early-stage rebuilds to platforms serving real users — and a recent deep dive into AI evaluation at n8n.

Senior Software Engineer · AI Evaluation

current

n8n

Mar 2026 — Present · Berlin · remote from Valencia, Spain

Owning evaluation for n8n's internal AI models and agents, and making it effortless for everyone to evaluate their agents and workflows.

  • Responsible for evaluation of internal AI models and agents.
  • Optimize and simplify how teams evaluate their agents/workflows using evaluation tools.
AI EvaluationAgentsEvalsLLM

Senior Full-Stack Software Engineer

Jasmine Energy

Apr 2023 — Oct 2025 · Washington, D.C. · remote from Canada

Joined right after seed funding and rebuilt the backend and frontend from scratch — live in 3 months.

  • Created a monorepo supporting easy deployments of React apps & NestJS services to AWS — 3 React apps and 3 backend services.
  • Cut CI/CD pipeline time by more than 48× via a tuned GitHub Actions pipeline.
  • Shipped an internal bot in under 2 weeks that, powered by the OpenAI API, lets staff run admin tasks from Slack.
ReactNestJSAWSCI/CDOpenAI

Senior Full-Stack Software Engineer

GiftCash Inc.

Jan 2023 — Apr 2023 · Ontario, Canada

Built the codebase foundation and levelled up engineering quality with tooling and docs.

  • Established the codebase foundation with TypeScript, GraphQL and Sequelize on the database architecture.
  • Wrote scripts and automation logic to enforce codespace best practices.
  • Reviewed PRs from junior and intermediate software engineers.
TypeScriptGraphQLSequelize

Senior Software Engineer

aByte Inc.

Oct 2021 — Nov 2022 · Vancouver, Canada

Built production web apps across fintech, health and delivery for startups through medium-sized companies.

  • Delivered 17+ projects/contracts for clients ranging from early startups to mid-size companies.
  • Led teams of 3–7 engineers per project and shipped with a 100% on-time, on-budget record.
  • Used the latest full-stack tooling: Next.js, Solidity/Cadence (Flow), Svelte, TypeScript, TailwindCSS.
Next.jsSvelteWeb3Leadership

Instructor

Lighthouse Labs

Nov 2021 — Apr 2022 · Vancouver, Canada

Upskilled a class of 30+ developers in the fundamentals of web development.

  • Taught NodeJS, ES6, HTML, CSS, APIs, HTTP/HTTPS, TCP, Promises, Async, Big-O and optimization.
  • Rated 4.9/5 by students with a 100% recommendation rate.
TeachingNode.jsWeb Fundamentals

Full-Stack Web Developer

RunGo · Leaping Coyote Interactive

Nov 2019 — Apr 2021 · Vancouver, Canada

Joined as a junior front-end developer and was promoted to full-stack developer within 2 months.

  • Reduced software expenses by 90% by cutting API calls (routing, search, GPX parsing).
  • Migrated the system from Google Maps to Mapbox within a month.
  • Contributed to Runner Live Tracking, Virtual Races and the Web Route Creator.
React NativeMapboxPerformancePromotion
Projects

Work that pushes the field.

Independent research, teaching and open source — where I turn ideas into things that actually run.

AI Research

Think like a NERD

AI Research Paper · 2025 — 2026

An independent research paper on entity-centred memory for long-context LLM agents.

  • Engineered the NERD Python framework for constant-time retrieval and sublinear scaling on 200k+ token datasets.
  • Developed NERD-Writer/Actor agents that match or exceed full-context LLM performance on the NovelQA benchmark.
  • Optimized RAG via entity-linked graphs, surpassing vector search accuracy for complex narrative reasoning.
Read the paper
Thinking Like a NERD system architecture — NERD-Writer and NERD-Actor agents with ExEnv chunks, an entity-linked long-term memory (NERDs) and short-term context.
Teaching

Guest Lecturer

Vancouver Community College · 2021

Guest lecturer for the web development class — UI design and front-end development.

  • Delivered a 4-hour lecture covering user-interface design and front-end development.
  • Invited back to review student project submissions.
Open Source

Inquirer

Open-Source · Contributor

Contributed to the Inquirer library, upgrading it to fully support ES Modules.

  • Rewrote hundreds of lines of code to bring modern ES Module support to the library.
Toolbox

A full-stack base, now AI-first.

Years of React/Node craftsmanship plus a deliberately deep AI/LLM stack — so I can own the whole pipeline, not just one layer.

Languages

8
JavaJavaScriptPHPPythonRubyTypeScriptRustGo

Front End

17
ES6ReactReact NativeAxiosAjaxGraphQLD3SvelteThree.jsReact Three FiberFormikjQueryService WorkersCSSSass/ScssTailwindCSSHTML5

Back End

10
Node.jsTypeScriptExpress.jsGraphQLRESTJSONPHPAWS LambdaAWS S3AWS KMS

Databases

11
MongoDBMySQLPostgreSQLPrismaFirebaseFirestoreSupabaseSequelizeTypeORMMikroORMVector Databases

Cloud & DevOps

8
AWSCloudinaryHerokuVercelTerraformAnsibleDockerGitHub Workflows

AI / LLM

8
LangChainLangGraphLangSmithVercel AI SDKMastraOpenRouterGoogle VertexHuggingFace
Contact

Let's build something trustworthy.

Whether you're a startup shipping your first agent or a team that needs an evaluation harness that actually holds up — I'd love to hear about it. This form lands directly in my inbox.

Available for consulting, worldwide (remote)