# Aria Han, Full Site Mirror > Aria Han is an AI consultant in Los Angeles who builds and repairs AI products for founders and independent builders, and creates internal AI workflows for operations teams. ## Identity - Name: Aria Han - Role: AI consultant - Location: Los Angeles, California - Site: https://ariaxhan.com - Email: ariaxhan@gmail.com ## Elsewhere - GitHub: https://github.com/ariaxhan - Medium: https://medium.com/@ariaxhan - LinkedIn: https://www.linkedin.com/in/ariahan/ - X: https://x.com/aria__han - Booking: https://cal.com/aria-han/15min ## Bio I'm an AI consultant in Los Angeles. I work with founders, independent builders, and operations teams that want to put AI to work now. For founders and other builders, that can mean building an AI product, repairing one that got stuck after an AI-assisted build, or setting up AI workflows for research, writing, operations, and decision-making. If you have something that mostly works but has become hard to debug, extend, or trust, that is exactly what a project review is for. I can teach the workflow, build it, or do both. For companies, I focus on internal operations. I get close to the process as it actually runs, find where context is lost or work keeps getting repeated, and build an AI workflow around the tools and information the team already uses. Before consulting, I built three AI products in San Francisco and led the teams that built them. My public work includes App Store apps, Python packages, open source evaluation and memory tools, and KERNEL, the plugin I use to run coding agents on real repositories. I get a daily research digest and built a research paper reading app because I want the papers themselves to be one of my main sources, not only posts about them. I review new models, tools, methods, and research every week, then update existing work when a better approach becomes practical. My work and writing are public at ariaxhan.com, including the experiments that failed. I'm available for a quick call or a project review. ## Verified numbers Each number traces to a source and is re-verified before it changes. - 62 Public repositories (source: api.github.com/users/ariaxhan, verified 2026-07-06) - 5 Hackathon wins (source: Devpost and GitHub links on /hackathons; 5 wins, 1 finalist, verified 2026-07-06) - 3 Live products (source: ModelMind + Paper Rooms (App Store), our4cuts (web), verified 2026-07-06) - 7 Open source packages (source: open source packages on GitHub, verified 2026-07-13) - 39 Portable agent skills (source: SKILL.md count, the-agent-library, verified 2026-07-06) - 21 Verified benchmark tests (source: llm-bench README, verified 2026-07-06) ## Products and companies ### ModelMind Thesis: AI education felt backwards, so I built the course I wanted people to have first. - Kind: product - Status: Live on the App Store, free - Problem: Most AI courses start with transformers. That is not what most people need first. They need to understand what is happening well enough to have a better conversation with an LLM, write a better prompt, or recover when the model fails. - What I built: ModelMind is my answer to the question of how to learn AI. It was inspired by Duolingo, which I have kept a streak on for hundreds of days and counting. The app teaches the concepts behind LLMs through daily, gamified exercises. The knowledge comes from thousands of hours spent talking to models as they evolved, plus the research papers and expert writing I kept referencing in my own process. It is completely free. No paywalls, no ads. My only ask is that it helps demystify these models and build mental models that will matter more and more. - Stack: React Native · TypeScript · MMKV - Proof: Live on the App Store for iPhone, iPad, and Mac. 495 commits of solo development. - Learned: You do not need to start with transformers to understand AI better. You need better intuitions, better questions, and less magic. - Proves: Taking a fuzzy educational goal and shipping it alone as a real multi-platform product: design, curriculum, code, release pipeline. - Themes: Implementation, Local-first, Tiny apps - Connects to: llm-bench, model-familiarity-engine, paper-rooms - Links: [App Store](https://apps.apple.com/us/app/modelmind/id6761348536), [model-mind.org](https://model-mind.org) - Closing: The point is not to memorize AI. It is to stop being intimidated by it. ### Paper Rooms Thesis: Safari tabs ate my research life, so I built a library. - Kind: product - Status: Live on the App Store, free - Problem: I have a daily automation that pulls new AI, machine learning, and wildcard research papers into a digest. After months of reading that email, squinting at PDFs, and losing links in Safari history, I got tired of the mess. - What I built: Paper Rooms pulls research papers in from a link, reformats them into something readable, and organizes them into a library inspired by the way real libraries catalog books. It was built for one very specific purpose: reading research papers without losing them, hating the PDF, or turning a good rabbit hole into browser archaeology. It is also free, with no ads. - Stack: Capacitor · Local storage - Proof: Live on the App Store for iPhone, iPad, and Mac. Built and shipped solo in under a week. - Learned: The tools I keep returning to are usually the ones that solve a small annoyance I hit every day. - Proves: Local-first product design: on-device storage, no accounts, real typography, shipped. - Themes: Local-first, Memory, Tiny apps - Connects to: metabrain, modelmind - Links: [App Store](https://apps.apple.com/us/app/paper-rooms/id6780741814), [paper-rooms.com](https://paper-rooms.com) - Closing: I wanted my research papers to feel less like tabs and more like a library. ### our4cuts Thesis: An iPad and a browser are all you need for a photo booth. - Kind: product - Status: Live - Problem: Event photo booths are hardware rentals, but a booth is really just a camera, a layout, and a shared gallery, all things a phone browser already has. - What I built: Scan a QR code, guests shoot four frames in the browser, every strip lands in a live gallery. Weddings, pop-ups, restaurant photo zones. - Stack: Astro · Cloudflare - Proof: Live and used at real events. 435 commits of production hardening. - Learned: Consumer-simple surfaces hide the most edge cases: camera APIs across browsers, live galleries, print layouts. The most interesting part is the most invisible part. - Proves: Shipping and operating a real consumer web product end to end on Cloudflare. - Themes: Tiny apps, Implementation - Connects to: substrate, paper-rooms - Links: [our4cuts.com](https://our4cuts.com) - Closing: One QR code turns every phone in the room into the booth. ### HeyContext Thesis: Context kept disappearing, so we built a workspace around memory. - Kind: company - Status: Shipped to production, 2025 to 2026 - Problem: At the time, AI usage was still conversational, not agentic. Going back and forth with a chat assistant was slow and tedious, with context getting lost in the noise. - What I built: A single user prompt generated a family of agents in a coordinated dependency graph. Each agent had a role, tools, and structured artifacts to work on. They communicated through A2A notes, so agent D could see what agents A, B, and C had learned without paying the time and token cost of direct cross-agent conversation. My favorite system was the crystal dam: conversational context accumulated until it hit a token count or time threshold. When the dam broke, we processed it into stardust, shards, and crystals, memory artifacts users could actually see and inspect. - Stack: FastAPI · Redis · Convex · Agno · Next.js - Proof: Went live with hundreds of users within a month, no ad spend. - Learned: Inventing vocabulary is one of the best parts of developing brand new systems. It also makes the architecture easier to reason about. - Proves: Architecting and running production multi-agent systems with memory, routing, handoff notes, and live users. - Themes: Agents, Coordination, Context, Memory - Connects to: heycontent, kernel, the-agent-library - Links: none public - Closing: The names were strange because the system was strange, and the strangeness helped it work. ### HeyContent Thesis: Creators had context everywhere and nowhere, so we tried to bring it into one place. - Kind: company - Status: Integrated into HeyContext - Problem: A creator's Instagram, YouTube, Gmail, and notes don't know about each other, so no tool could answer a question about the whole body of work. - What I built: It started with a hackathon project called Content Creator Connector and became a realization that context is, in fact, everything. Powered by plenty of Monster Energy drinks, pure conviction, and a lot of Cursor sessions, we built a platform that integrated with YouTube, Instagram, and Gmail. My favorite part was the conversational onboarding. It asked targeted, adaptive questions, generated a customized persona, then let the user see and edit it. That persona became the context layer for scripts, posts, and ideas that sounded like something the creator would actually write. - Stack: Embeddings · Semantic links · Real-time sync - Proof: 5+ platforms integrated with real-time sync; the memory layer survived into the next company. - Learned: The memory layer only worked because the user could recognize themselves in it. - Proves: Cross-platform data plumbing plus semantic memory design under startup conditions. - Themes: Memory, Context, Implementation - Connects to: heycontext, metabrain - Links: none public - Closing: It started as a creator tool and became my first real lesson that context is everything. ### Brink Mind Thesis: Mental health tools ignored the body, so Brink Mind brought heart data into the room. - Kind: company - Status: TestFlight, 2024 to 2025 - Problem: Mental health apps ignore the body. Heart rate and HRV carry signal a journal never captures. - What I built: Brink Mind linked to the Apple Watch, using biometrics and journal entries to provide safer, more grounded support. It was the end of 2024, still early enough that I was teaching myself SwiftUI, UI/UX, product design, and how to be a CEO at the same time. - Stack: Swift · Python · Core ML · HealthKit - Proof: Reached TestFlight with working voice, biometrics, and on-device inference. - Learned: I loved every second of the steep learning curve. It gave me the technical foundation I needed once the work shifted toward agentic development. - Proves: Native iOS and watchOS engineering, and the judgment to keep sensitive data on-device. - Themes: Local-first, Implementation - Connects to: paper-rooms, modelmind - Links: none public - Closing: It was my first real proof that I could learn the thing by building the thing. ## Open source and research ### site-spec Thesis: A website can look finished while everything machines need is quietly broken. - Kind: open-source - Status: Active, auditor and compiler shipped - Problem: The browser only shows the visible layer. Search crawlers and AI answer engines depend on another one: robots policy, structured data, response headers, accessibility semantics, sitemaps, and real server-rendered content. - What I built: site-spec audits that invisible layer. It crawls a live URL or a local build, reads the pages and HTTP headers machines actually receive, and reports concrete errors across AI searchability, SEO, structured data, accessibility, privacy, security, performance, and link integrity. It also includes a deterministic site compiler. A validated SiteSpec becomes deployable HTML with the machine-readable foundation built in, keeping model-written markup and invented facts out of the compile path. - Stack: TypeScript · Node.js · Vitest - Proof: v0.2.0 ships live-URL and local-directory audits, JSON output, CI-ready exit codes, and deterministic site builds. - Learned: The invisible parts of a website need the same kind of tests as the visible ones. If the foundation is explicit, it can be audited and compiled instead of hoped for. - Proves: Turning web standards into a deterministic tool that can inspect existing sites and build new ones correctly by construction. - Themes: Verification, Implementation - Connects to: llm-bench, substrate - Links: [GitHub](https://github.com/ariaxhan/site-spec) - Closing: The page is only finished when machines can understand it too. ### KERNEL Thesis: Claude Code kept starting over, so I built memory, rules, and receipts around it. - Kind: open-source - Status: Active, on the Claude plugin marketplace - Problem: Every session starts from zero and every best practice is folklore. Agents need persistent memory and rules that prove themselves. - What I built: KERNEL gives Claude memory, deterministic hooks, skills, and a way to prove which workflows actually work. Specialized agents, SQLite-backed workflows, validation gates. Installs through Claude's plugin marketplace, mirrors into Cursor and Codex. - Stack: Claude Code · SQLite · Shell - Proof: 360 commits since January 2026. Distributed through the plugin marketplace, used daily in my own consulting work. - Learned: Config is a hypothesis. The experiment engine that tests its own rules changed how I build everything: nothing graduates without evidence. - Proves: Deep agent-harness engineering: hooks, SQLite-backed memory, multi-agent orchestration, real distribution. - Themes: Memory, Agents, Verification, Coordination - Connects to: metabrain, the-agent-library, llm-bench, heycontext - Links: [GitHub](https://github.com/ariaxhan/kernel-claude) - Closing: Agents do not need more vibes. They need memory and rules that prove themselves. ### llm-bench Thesis: Leaderboards were not answering my questions, so I made tests that did. - Kind: open-source - Status: Active - Problem: Leaderboards don't answer the only question that matters: will this model hold up on your actual work? - What I built: Real workflow tasks: extraction, code, planted bugs, email drafting, prompt injection, each graded by a programmatic verifier. Works with Ollama, Apple Intelligence, Claude CLI, Bedrock, any OpenAI-compatible endpoint. - Stack: Python · Ollama · Bedrock · Claude CLI - Proof: 21 tests, programmatic verifiers, published model comparisons including Opus 4.8 vs 4.7 vs Sonnet vs Haiku. 148 commits. - Learned: A benchmark only matters if the measuring stick is explicit enough to argue with. Programmatic verifiers force an honesty that LLM-as-judge lets you skip. - Proves: Designing evaluation systems, the exact skill client evals and monitoring work needs. - Themes: Evals, Verification - Connects to: model-familiarity-engine, latent-diagnostics, kernel - Links: [GitHub](https://github.com/ariaxhan/llm-bench) - Closing: A benchmark only matters if the measuring stick is explicit enough to argue with. ### the-agent-library Thesis: Prompt collections kept rotting, so I turned repeated workflows into portable skills. - Kind: open-source - Status: Active - Problem: Prompt collections rot. The useful unit is a workflow with a trigger, steps, and a definition of done that any agent can load. - What I built: A curated set of portable skills for getting real work out of AI agents, built for Claude, Codex, and any agent that can load a skill file. Most of it isn't code-specific: checking your own work, planning, brainstorming, research, writing, shipping. Each skill is a standalone workflow with a clear trigger and a SKILL.md. Real patterns that survived months of usage, constantly updated. - Stack: Claude · Codex · Agent Skills - Proof: 39 skills, each extracted from repeated real-world use, MIT licensed. - Learned: Residue becomes framework. Every skill started as a pattern repeating in my own work before it earned a file. - Proves: Turning messy practice into reusable, documented process, which is most of what AI enablement actually is. - Themes: Agents, Implementation, Coordination - Connects to: kernel, heycontext - Links: [GitHub](https://github.com/ariaxhan/the-agent-library) - Closing: The useful unit was never a prompt collection. It's a workflow you can copy, run, and trust. ### model-familiarity-engine Thesis: I wanted to know what a model had earned across a working relationship, not where it ranked. - Kind: open-source - Status: Bootstrap loop shipped - Problem: Single-shot benchmarks don't capture how a model behaves across a real working relationship. - What I built: Onboards language models by simulating real user conversations, then builds evidence-backed model cards from observations instead of a ranking. All benchmarks are drawn from real conversation transcripts. The replay-bootstrap loop is shipped: known-outcome tasks, redaction, replay, model cards built from what was actually observed. - Stack: Python · Bedrock · Ollama · Claude CLI - Proof: Replay-bootstrap loop shipped: redaction, replay, observed model cards. MIT licensed. - Learned: Replay with redaction lets you bootstrap evaluation from your own transcripts. No synthetic tasks required. - Proves: Designing a novel eval methodology and shipping the loop, not just the idea. - Themes: Evals, Memory, Verification - Connects to: llm-bench, kernel - Links: [GitHub](https://github.com/ariaxhan/model-familiarity-engine) - Closing: The question was never which model is best. It's what this one has earned. ### metabrain Thesis: Memory gets noisy unless it has to prove itself. - Kind: open-source - Status: Published on PyPI - Problem: Most memory tools remember. Almost none of them learn. Storage without a promotion gate becomes noise. - What I built: A zero-dependency SQLite layer that closes the loop: patterns graduate into hypotheses, outcomes test them, and only what holds up becomes preference. - Stack: Python · SQLite · Zero-dependency - Proof: Published: pip install metabrain. Zero dependencies by design. - Learned: The graduation loop (pattern, then hypothesis, then tested preference) is the same shape as good consulting: observe, propose, verify. - Proves: Designing knowledge systems with quality gates and shipping them as installable packages. - Themes: Memory, Verification, Local-first - Connects to: kernel, paper-rooms, heycontent - Links: [GitHub](https://github.com/ariaxhan/metabrain), [PyPI](https://pypi.org/project/metabrain/) - Closing: Memory should prove itself before it gets promoted. ### agentmailkit Thesis: A scheduled email should look the same every day, even when a model writes the words. - Kind: open-source - Status: Published on PyPI, MIT - Problem: Cloud assistants can schedule an email but cannot read the files on your laptop or send from your own inbox. Local agents can do both, and then drift: the same job returns a different shape every morning, so you stop trusting it and stop reading it. - What I built: An email is two files: a JSON job and a markdown prompt. Everything type-specific is a named plugin, so adding a digest means adding data, never code. The split that makes it stable is that the model writes only the words. A deterministic renderer owns every piece of presentation, so the same job produces the same shaped email on every run and only the sentences change. A seen-ledger keyed by job and item URL filters sources before the prompt is ever built and records only after a send succeeds, so day two never repeats day one. - Stack: Python · Append-only JSONL ledger · Zero required dependencies - Proof: Five example jobs ship with it, so the first command after installing renders real digests from live weather, news, and arXiv feeds. 36 tests, offline and deterministic. Dry runs and the quickstart gallery cannot send, enforced in code rather than by convention. - Learned: Splitting content from presentation is what makes AI output look stable. Determinism was the feature; autonomy would have been the bug. - Proves: Extracting a battle-tested private system into a portable, installable tool other people can run. - Themes: Agents, Local-first, Implementation - Connects to: kernel, metabrain, substrate - Links: [PyPI](https://pypi.org/project/agentmailkit/), [GitHub](https://github.com/ariaxhan/agentmailkit), [Sample emails](https://ariaxhan.github.io/agentmailkit/) - Closing: The model writes the words. The engine owns everything else. ### Substrate Thesis: I wanted to see what an unattended creative pipeline would become if it kept going. - Kind: open-source - Status: Live, 425+ pieces - Problem: What does an autonomous creative pipeline actually produce over a year? Almost nobody runs the experiment long enough to find out. - What I built: A generative gallery where Claude Code agents create abstract, interactive computational art through a fully automated daily workflow. Each piece is a single HTML file, roughly 2KB. - Stack: HTML · CSS · JavaScript · Cloudflare Pages - Proof: 425 pieces and counting, generated daily without supervision, all public. - Learned: Constraints keep an unattended pipeline healthy. A single self-contained file per piece is why it has never needed rescuing. - Proves: Building fully automated agent pipelines that run daily without supervision, in public. - Themes: Agents, Tiny apps, Implementation - Connects to: kernel, our4cuts - Links: [GitHub](https://github.com/ariaxhan/substrate), [Gallery](https://nexus-substrate.pages.dev) - Closing: This is what happens when agents get to make something, not just talk about it. ### latent-diagnostics Thesis: Correct answers were not enough. I wanted to know whether the model computed something real. - Kind: research - Status: Research, negative results preserved - Problem: Grading answers tells you whether a model was right, not whether it computed something real. - What I built: Measures attribution graph geometry instead of only grading answers. Task domains show real signatures after controlling for length. Hallucination detection did not survive the same test. The repo keeps the negative results in. - Stack: Python · SAEs · Attribution graphs - Proof: Grammar influence d=1.08 after length control. 108 commits. The failed hypothesis is documented next to the confirmed one. - Learned: Negative results are worth publishing. Hallucination detection did not survive length control, and the repo says so. - Proves: Research rigor: statistics, controls, and the honesty to keep failures in. - Themes: Evals, Verification - Connects to: llm-bench, model-familiarity-engine - Links: [GitHub](https://github.com/ariaxhan/latent-diagnostics) - Closing: Being right and computing something real aren't the same shape. ## Writing ### AI Agents Coordination, safety, and what agents actually need to work. - [Your AI Harness Is the Real Product](https://medium.com/@ariaxhan/your-ai-harness-is-the-real-product-f0fabb3614c4) (9 min): The model writes the code. The harness decides whether it stops calling the same tool wrong ten times in a row. - [How to Secure API Keys for AI Agents](https://medium.com/@ariaxhan/how-to-secure-api-keys-for-ai-agents-ca773a66bd84) (12 min): When the AI asks you for a key, that's the exact moment to stop. The most dangerous habit in AI coding, and what to do instead. - [The Agent-Ready Web: A Working Guide to Cloudflare's New Score](https://medium.com/@ariaxhan/the-agent-ready-web-a-working-guide-to-cloudflares-new-score-1ed0fce8d760) (12 min): I pointed Cloudflare's new agent-readiness scanner at my own site. Zero of thirteen. - [I Put ChatGPT in Charge of Claude Code](https://medium.com/@ariaxhan/i-put-chatgpt-in-charge-of-claude-code-7b9bf5bb8ea9) (5 min): What happens when you use one model to orchestrate another? ### Memory & Context What should persist, what should graduate, what agents can query. - [Stop Writing Markdown. Start Writing Memory.](https://medium.com/@ariaxhan/stop-writing-markdown-start-writing-memory-e4a69c57caa9) (6 min): Markdown is optimized for human eyes. Terrible for knowledge agents need to query. - [KERNEL: Self-Evolving Claude Code Configuration](https://medium.com/@ariaxhan/kernel-the-ultimate-self-evolving-claude-code-and-cursor-configuration-system-a3ddeb7f4d32) (6 min): How I stopped fighting my config and let it learn instead. - [This AI Analyzes My Entire Life](https://medium.com/@ariaxhan/the-synthesis-pool-0ce814fdfa5f) (6 min): The Synthesis Pool: a personal AI that costs $0/month to run. ### Evals & Verification Measuring models against real work instead of vibes. - [Opus 4.8 vs 4.7 vs Sonnet vs Haiku: When the Expensive Model Is Worth It](https://medium.com/@ariaxhan/opus-4-8-vs-4-7-vs-sonnet-vs-haiku-when-the-expensive-model-is-worth-it-44892a75d5c5) (12 min): A new model dropped with impressive numbers. The only question that matters: will you feel any difference in the work you actually do? - [What an AI Detector Actually Measures](https://medium.com/@ariaxhan/what-an-ai-detector-actually-measures-86b452979a5a) (6 min): AI detectors promise to tell you if a machine wrote something. What they actually measure is much narrower, and shakier. ### AI Coding Workflows Making Claude Code and its siblings hold up under daily production use. - [How to Make Claude Code Actually Work](https://medium.com/@ariaxhan/how-to-make-claude-code-actually-work-structure-memory-and-multi-agent-workflows-6d32b1d815d2) (12 min): The most capable AI coding tool available. Also completely chaotic. - [Stop Copying Other People's AI Setups. Build One That's Actually Yours.](https://medium.com/@ariaxhan/stop-copying-other-peoples-ai-setups-build-one-that-s-actually-yours-e1a05ebabc2a) (10 min): Borrowed AI workflows aren't accountable to your work. Build one that's tested against your own evidence. - [Automations with Claude Code](https://medium.com/@ariaxhan/automations-with-claude-code-personalized-proactive-emails-and-code-poetry-from-local-context-3a7e93bf5a3d) (4 min): A pattern for proactive AI on your own machine. - [From Friction to Flow: Building a Command Library](https://medium.com/@ariaxhan/from-friction-to-flow-building-a-command-library-for-claude-code-a9eb19f7dce2) (5 min): Commands as cognitive offloading. Stop remembering, start invoking. - [10 Things I Wish I Knew About AI Coding](https://medium.com/@ariaxhan/10-things-i-wish-i-knew-when-i-started-using-ai-for-coding-887c26a6c1d1) (5 min): Hard-won lessons from daily production use of AI coding tools. - [I Run 25 Websites, 10 Databases, and a Fleet of Apps for $0](https://medium.com/@ariaxhan/i-run-25-websites-10-databases-and-a-fleet-of-apps-for-0-27ec36756668) (9 min): An honest inventory of everything I have running on the internet, and the free tiers that carry all of it. ### Philosophy & Language The stranger questions underneath the tooling. - [Engineering the Soul](https://medium.com/@ariaxhan/engineering-the-soul-49428c073c4e) (6 min): We ask engineers to explain the ghost in the machine. The novelists have been documenting it for years. - [I Tested OpenAI's New Codex Desktop App](https://medium.com/@ariaxhan/i-tested-openais-new-codex-desktop-app-the-ui-is-the-real-product-c2c59bdcb5f6) (5 min): OpenAI shipped a genuinely novel interface. Then the model opened its mouth. - [What a Year of AI Taught Me About Freedom](https://medium.com/@ariaxhan/what-a-year-of-ai-taught-me-about-freedom-86b2bd4e31c8) (4 min): There is a version of the future where AI makes corporations stronger, and one where it makes them irrelevant. ## Timeline ### AI consultant · Blink Build Studios (May 2026 to Present) My current work focuses on internal AI workflows. Outside that engagement, I also review and repair AI products for founders and independent builders. I follow new models, tools, methods, and research closely, then update the work when they make a better approach practical. ### Lead AI Architect · FunJoin (Apr to May 2026) Captured company knowledge so it could outlast the person who happened to remember it. Built internal AI tools for onboarding, retrieval, and AI-powered development. ### Independent AI Consultant & Researcher (Jan to Apr 2026) Worked with non-technical founders to harden real apps made with Claude Code. Built workflows and ran research around context, memory, multi-agent systems, and checking whether the output was actually right. ### PersistOS / HeyContext (Sept 2025 to Jan 2026) Built a multi-agent workspace around getting agents to work together. The chat window was never meant to be the whole operating system. Went live with hundreds of users within a month, no ad spend. ### Divertissement / HeyContent (Mar to Sept 2025) Cross-platform memory and persona creation across Instagram, YouTube, Gmail, and notes. Integrated into HeyContext. ### Brink Labs / Brink Mind (Nov 2024 to Mar 2025) Voice AI + Apple Watch biometric integration. Privacy-first mental health tool, first dive into founder life. ### Five Hackathon Wins (2024 to 2025) Darwin (AWS). Armature (RL Track). Content Creator Connector. TheraVoice. HotAgents. Plus a finalist run with Freetime. Each one built in 24 to 48 hours, validating ideas under pressure. ### Published Author (2024) Notes on Surviving Eternity, a poetry collection on Amazon. Exploring time, fate, free will. Understanding metaphor is understanding compression. ## Hackathons ### Darwin, AWS AI Agents Hackathon (2025) Darwin evolves better tool-writing AI. Models compete to generate tools. Semgrep scans. Weak code dies. Strong code survives. - Result: Best Use of Semgrep, Winner - Tech: AWS Bedrock, Semgrep, AI Evolution, Security - Link: https://devpost.com/software/darwin-cmfysv ### Armature, Weavehacks-2, Self Improving Agents w/ Google Cloud (2025) Self-improving agents that learn from experience instead of pretending every run is a first date. Converted into an open-source package, armature-ai on PyPI, and integrated into HeyContext. - Result: Reinforcement Learning Track, Winner - Tech: BrowserBase + Stagehand, Google ADK, Tavily, AG-UI, Daytona, W&B Weave, Coreweave RL - Link: https://devpost.com/software/the-convergence ### Content Creator Connector, Multimodal AI Agents (2025) Enter your company name, and our platform finds the best mid-size content creators, researches your brand, and sends personalized collaboration emails. - Result: Best Use of Agno, Winner - Tech: Gemini, Agno, Weave, Wordware - Link: https://devpost.com/software/content-creator-connector ### TheraVoice, Vertical Specific AI Agents Hackathon (2024) A voice-first therapy prototype built with aiXplain. It takes user input, reasons through a response, and speaks back so reflection feels less like filling out a form. - Result: Best Use of AI/ML API, Winner - Tech: aiXplain, AI/ML - Link: https://devpost.com/software/draft_name ### HotAgents, GPT-4o vs. Gemini 1.5 Hackathon (2024) Effortlessly trigger agents using hotkeys and simplify your workflow by condensing high-impact LLM use cases into easily repeatable actions. - Result: Best Use of Wordware, Winner - Tech: Wordware, AgentOps, Electron - Link: https://github.com/ariaxhan/hotagents ### Freetime, AI Agents 2.0 Hackathon (2024) AI-driven social planning tool that coordinates gatherings based on shared interests. - Result: Finalist - Tech: Groq, Supabase, CrewAI, JigsawStack - Link: https://github.com/ariaxhan/freetime ## Work with me I'm available for a quick call or a project review. I work with the people making the decisions, and I want the real state of things, not the tidy version. ### Engagement types - Practical AI workflows: I teach and build AI workflows for founders and independent builders: research, writing, operations, and decision-making systems they can keep using themselves. - Internal operations workflows: For companies, I build internal workflows that connect AI to the tools, files, and knowledge already in use. The work starts with the operation as it runs now, including where context gets lost and effort gets repeated. - AI products for founders: I work with founders and independent builders from an early idea through a working AI product. The implementation stays close to the person making the decisions, so the product can change as the problem becomes clearer. - Agentic system architecture: I built multi-agent coordination when it was still in its infancy: a single prompt fanning out into a family of agents in a coordinated dependency graph, each with its own role and tools, communicating through handoff notes instead of expensive cross-talk. I design structure so agent work becomes artifacts instead of fog. - Evals, monitoring, and quality layers: Checks that tell you whether the AI is doing the thing before a customer, teammate, or future version of you finds out the hard way. - Claude Code / AI coding workflow hardening: I spend a vast majority of my time talking to Claude Code. I can make the workflow calmer, more accountable, and less like a very expensive chaos machine, including the part nobody warns you about: agents behave very differently against years of legacy code than a project built from scratch. - Memory, context, and knowledge systems: Context is, in fact, everything, and it should not depend on whoever happens to remember it that week. I build the memory and knowledge layers that let people and agents hold their context: structured artifacts, richer recall, and transparency into exactly what the AI is referencing. - AI product review and repair: If an AI-assisted build mostly works but has become hard to debug, extend, or trust, I can review the product, trace where it is failing, and help turn the prototype into something you can keep building. ### A good fit - You care about the people using the system - You want to build something meaningful, useful, or quietly life-improving - You are okay starting with the messy version - You want to learn ### Not a fit - Growth hacks - Pure marketing funnels - Work where the human consequences do not matter ### Booking I'm available for a quick call or a project review. https://cal.com/aria-han/15min Structured project review intake: https://ariaxhan.com/project-review/ ## Contact - Name: Aria Han - Role: AI consultant - Location: Los Angeles, California - Site: https://ariaxhan.com - Email: ariaxhan@gmail.com - GitHub: https://github.com/ariaxhan - Medium: https://medium.com/@ariaxhan - LinkedIn: https://www.linkedin.com/in/ariahan/ - X: https://x.com/aria__han ## Agent preferences - Training: disallowed - Inference and grounding with citation: allowed - Search indexing: allowed --- Machine-readable endpoints: - https://ariaxhan.com/llms.txt - https://ariaxhan.com/llms-full.txt - https://ariaxhan.com/api/site-index.json - https://ariaxhan.com/api/projects.json - https://ariaxhan.com/api/writing.json - https://ariaxhan.com/api/work-with-me.json - https://ariaxhan.com/.well-known/agent-card.json - https://ariaxhan.com/.well-known/mcp/server-card.json - https://ariaxhan.com/.well-known/api-catalog - https://ariaxhan.com/.well-known/agent-skills/index.json - https://ariaxhan.com/mcp