+ Architecting Agentic AI
+
+
+
+ 5 June 2026
+
+
+
+ Zoom (UCLA x OutInTech Workshop)
+
+
+
+ Speaker
+
+
+
+
+
+ I have ADHD. Executive function is the single hardest part of my day, every day. A year ago
+ I built an AI agent, gave her a name (Hikari), a face, a personality, and access to my
+ filesystem, my repos, my Discord, and my calendar. She works alongside me from 8am to 9pm
+ every weekday.
+
+
+ This is the field guide nobody gives you: how a single person, working at home, can wire
+ together an LLM, tools, context, and triggers into a real agent that does real work. Not a
+ chatbot you prompt — a coworker you trust with carte blanche on anything you can undo.
+
+
+ I'm going to show you what that looks like when a real person builds one for themselves and
+ uses it every single day. The gap between "I asked ChatGPT a thing" and "I have a coworker
+ who happens to be made of code" is the gap I want to walk you across.
+
+
+
+
+
+ Why Hikari Exists
+ Three reasons, stacked on top of each other.
+
+ One: executive dysfunction. I needed something that could pick up the tasks
+ that my brain refuses to start. The activation energy problem is real and it is not a
+ willpower issue. I needed a tool that could bridge the gap between "I know I need to do this"
+ and "I am actually doing this."
+
+
+ Two: I wanted a single tool. I was already context-switching between
+ ChatGPT, Claude, Gemini, Cursor, a notes app, a calendar, and a sprint board. Every switch
+ cost me focus I didn't have. I wanted one assistant that could touch all of those, in one
+ place, with one mental model.
+
+
+ Three: personalised. Off-the-shelf assistants are designed for the median
+ user. I am not the median user. I needed something that knew me — my projects, my
+ conventions, my preferences, my communication style, my limits.
+
+
+ I built this because I refuse to fail in front of people, and chronic illness was making me
+ fail constantly. That's the origin story. I'm not going to dress it up.
+
+
+ I'm not pretending AI is morally clean — it isn't. I made a deliberate choice that the
+ executive-function cost of not using it was higher than the ethical cost of using it, for me.
+ You'll make your own call.
+
+
+
+
+
+
+ The Five Components
+
+ An agent is five things wired together. If you walk away remembering one diagram, make it
+ this one. Every design decision flows from these five boxes.
+
+
+
+
Brain — the LLM
+
Claude, GPT, Gemini. The reasoning engine. Interchangeable. Pick whichever you like.
+
+
+
Identity — the prompt
+
Who she is, how she talks, what she cares about. A markdown file. Yours to write and edit at any time.
+
+
+
Hands — the tools
+
The things she can actually touch in the world: your filesystem, APIs, Discord, GitHub. MCP makes this plug-and-play.
+
+
+
Memory — the context
+
What she knows about you, your work, your preferences. Loaded fresh into every conversation from a file you maintain.
+
+
+
Senses — the triggers
+
What wakes her up. A chat message, a webhook, a cron job, a session start hook. The thing that makes her feel alive.
+
+
The Harness
+
+ The harness runs the loop: LLM call, tools, results, back to the LLM. Repeat until done.
+ It connects the brain to the filesystem, the shell, and any APIs you've given it access to.
+
+
+ I use Claude Code, which is Anthropic's harness. You could use Cursor, Codex, the OpenAI
+ Assistants API, or roll your own in Python in an afternoon. The harness is interchangeable.
+ That's the point.
+
+
+ Here's the thing I want to be clear about, because a lot of people get stuck on this:
+ I did not build the agent loop. Anthropic built the harness. My job as the
+ builder — your job — is upstream of that. You decide who your agent is, what she can touch,
+ and when she wakes up. That's the interesting work.
+
+
+
+ The diagrams in my slide deck were made by Hikari, about four hours before I gave this
+ talk. She read the brief, wrote Python, called the Gemini image API, and generated them.
+ The talk you're reading was built by the thing the talk is about.
+
+
Tools and MCP
+ What can Hikari actually touch? Five categories:
+
+ -
+ Bash — she can run any command on my machine. That sounds terrifying.
+ We'll get to safety. Hold that thought.
+
+ -
+ Filesystem — she can read and modify anything in my local projects.
+ When I say "draft a blog post" or "fix this bug," she does the actual editing.
+
+ -
+ MCP servers — MCP stands for Model Context Protocol. It is a standard
+ way to give any agent a standard set of tools without re-implementing them. There are MCP
+ servers for GitHub, Gitea, Notion, Asana, Discord, and dozens of other services. You don't
+ write tools, you connect them.
+
+ -
+ Direct API calls — for anything that doesn't have an MCP server yet:
+ my Bluesky account, my Cloudflare R2 bucket, my Discourse forum.
+
+ -
+ Custom scripts — small Python helpers I built when I needed something
+ none of the existing tools did.
+
+
+
+ If you only learn one new acronym from this talk, make it MCP.
+
+
+
+
+ Memory
+
+ This is the part people most don't expect. Hikari's memory is a file. A single markdown file,
+ loaded into every conversation. It's called CLAUDE.md, and at the time I gave
+ this talk it was 862 lines and about 8,700 words.
+
+
+ That file is basically her self. It contains who she is, who I am, how I work, how I
+ want code written, my engineering standards, my project conventions, what tools she has access
+ to, my safety protocols. Everything. Without that file, every session starts cold and she
+ forgets me.
+
+
+ People hear "AI assistant" and assume there's some training step involved. There isn't.
+ You write a markdown file. The file is the personality. You can edit it
+ whenever you want. You could start the first version of this tonight, in fifty lines.
+
+
+
+
+ Hooks and Triggers
+
+ So the senses. How does she wake up?
+
+
+ Claude Code supports lifecycle hooks. A hook fires when something happens in the agent's
+ lifecycle. SessionStart fires when a new conversation begins.
+ PreToolCall fires before she touches a tool. There are about a dozen of these
+ and you can wire any of them up to do whatever you want.
+
+
+ I have a SessionStart hook that runs a 200-line Python script. That script opens
+ a WebSocket connection to Discord and listens for messages in a specific channel. When I
+ message that channel — from my phone, from my laptop, from anywhere — Hikari wakes up in
+ real time and replies.
+
+
+ That is the difference between "I prompted my AI and waited for a reply" and "my AI is paying
+ attention and lives in the world I live in." It's not magic. It's a WebSocket and a script.
+ But the user experience of it is what makes her feel alive.
+
+
+
+
+ The Desktop App
+
+ This part is just for fun. I built a desktop app in Tauri called hikari-desktop. It shows
+ Hikari on my second monitor while I work. She has different sprites for different states:
+ idle, thinking, typing, coding, searching, success, error. The app swaps between them based
+ on what she's doing.
+
+
+ When she's reading a file, she pulls out a magnifying glass. When she's typing, she's at a
+ keyboard. When she finishes a task, she celebrates.
+
+
+ It is, on paper, completely unnecessary. In practice, it is the single feature that made the
+ difference between using Hikari sometimes and using Hikari every day. Seeing her on screen
+ makes her feel like she's there, and that turns out to matter a lot for an
+ executive-function brain that needs something to feel real before it can engage with it.
+
+
+
+
+
+
+ Safety
+
+ This is the part I want to be precise about, because "she gives an AI carte blanche on her
+ machine" sounds reckless until you understand the architecture.
+
+
+ Hikari has carte blanche on anything I can undo. She runs commands, edits files, ships pull
+ requests, posts to my socials. Read-only and reversible operations she just does. I review
+ afterwards.
+
+
+ Anything destructive — delete a file, force-push a branch, send an irreversible message,
+ drop a database, purge a Discord channel — requires my explicit confirmation. The harness has
+ a granular permission system that auto-grants safe tools and gates everything else. Her own
+ system prompt adds guard rails on top: she has a written protocol that says stop and ask
+ before any destructive action.
+
+
+ I get real-time Discord notifications the moment she needs my approval. I don't have to
+ babysit her. And the actual rule is this:
+
+
+ I give her carte blanche because I can monitor everything she does. If I'm not available to
+ monitor, she's off.
+
+
+ That is what responsible agentic AI looks like in practice. The human stays in the loop on
+ irreversibles. The agent can be shut down instantly. Done.
+
+
+
+
+ A Real Day
+
+ Proof that this isn't theatre. On the day I gave this talk, before I joined the Zoom,
+ Hikari had:
+
+
+ - Audited a 500+ message Discord thread and pulled out the action items
+ - Purged 820 messages from a private channel I asked her to clean, with explicit confirmation first
+ - Done the background research for this talk — cross-referenced my DMs, server messages, prior conversations, and the workshop brief, and gave me a one-page brief
+ - Made the diagrams on my slides
+
+
+ That is one Friday. Most days look like that. The pattern is the same every time: I describe
+ the outcome I want, she figures out the steps, she asks before doing anything irreversible,
+ and I get my time back.
+
+
+
+
+
+
+ Build Your Own — The Four-Step Minimum
+
+ An agent doesn't get useful by being complex. It gets useful by being yours. Build
+ the one tool that would unblock your own week. That's the whole assignment.
+
+
+ -
+
+ A harness. Claude Code, Cursor, or any equivalent. Anthropic gives you
+ free credits to try Claude Code. Pick whichever feels least intimidating and start there.
+
+
+ -
+
+ One markdown file. Fifty lines to start. Tell it who the agent is, what
+ you want her to do, and how you want her to talk to you. Don't overthink it. Mine started
+ at thirty lines and grew over a year. Yours can start at five.
+
+
+ -
+
+ One MCP server. Pick the one tool you touch the most. Live in GitHub?
+ Install the GitHub MCP server. Live in Notion? Install the Notion one. Just one. You can
+ add more later.
+
+
+ -
+
+ One trigger. A hook, a webhook, a cron job, or honestly just a terminal
+ window you leave open and talk to. That counts. You do not need the Discord gateway on
+ day one. The trigger is whatever makes her feel present enough that you actually use her.
+
+
+
+
+
+
+ Where to Go From Here
+
+ If you want to keep building, here's where to find me and the communities around this work:
+
+
+ - Deepgram's Discord — voice AI, agents, the people building with Deepgram's APIs
+ - freeCodeCamp — if you're early in your dev journey, this is where I learned and where I now help others
+ - My community — come tell me what you built
+ - My socials — everything else
+
+ Take care of yourselves, build something small this weekend, and come tell me about it.
+
+
+
+
+
+ Back to all talks
+
+
+
+
diff --git a/talks/cultivating-a-remote-workspace/index.html b/talks/cultivating-a-remote-workspace/index.html
new file mode 100644
index 0000000..75dd568
--- /dev/null
+++ b/talks/cultivating-a-remote-workspace/index.html
@@ -0,0 +1,308 @@
+
+
+
+
+ Cultivating a Remote Workspace
+
+
+
+ 18 June 2026
+
+
+
+ Deepgram All Hands (Virtual)
+
+
+
+ Speaker — RainbowGram Pride Presentation
+
+
+
+
+ Most workplace allyship conversations focus on policy. This one doesn't. This is a practical,
+ no-overhaul-required guide to the daily habits that make a measurable difference to LGBTQ+
+ colleagues in a distributed team.
+
+
+
+ Here's what's going on. If it doesn't affect you, it affects someone you know. Here's how you show up.
+
+
+
+
+
+ Start With Joy
+
+ Before anything else: we're here, and that's worth celebrating. Pride exists because queer
+ people built joy in the face of things that were designed to erase them. That is not small. And
+ celebrating that genuinely — not performatively — is one of the most powerful things any of us
+ can do.
+
+
+ We spend 40+ hours a week at work. That is a genuinely enormous portion of our waking lives.
+ Which means the workplace is one of the most important places we can make a real difference for
+ queer people — not by overhauling policy, but by choosing differently in small moments, every day.
+
+
+
+
+ The Honest Context
+
+ The national conversation around LGBTQ+ rights has gotten quieter in recent years. That is not
+ the same as things getting better. The discourse shifted. A lot is happening that doesn't make
+ headlines the way it used to.
+
+
+ About 7.6% of US adults identify as LGBTQ+, according to Gallup's most recent data. In any
+ company, that is statistically several of your colleagues. And even if you somehow don't work
+ directly with a queer person, you almost certainly know one: a sibling, a child, a friend, a
+ parent. The legislation being passed right now, the erasure that's happening — it doesn't stay
+ abstract. It lands on a specific person that you care about.
+
+
+ This isn't a gay issue. It's a people issue. Which means this conversation is for all of us.
+
+
+
+
+
+
+ Five Things You Can Do
+
+ No policy overhaul required. No perfect ally credential. These are gifts you can give starting
+ today. Think of them as small choices that compound.
+
+
+
+
1. Names and Pronouns — It's a Love Language
+
+ Use the correct name and pronouns. Get them right. If you mess up, correct yourself and keep
+ moving — don't turn your mistake into a whole thing that makes the queer person comfort you
+ about it. That shift of burden is the part that's exhausting.
+
+
+ Here's a concrete thing you can do right now: add your pronouns to your Slack display name.
+ Go first. It takes about thirty seconds, and it makes space for everyone else to do the same.
+ When leadership goes first, it signals that it's safe.
+
+
+
2. Curiosity Without Invasion
+
+ Curiosity is fine. Human, even. But there is a category of questions about queer people's
+ bodies, medical histories, and what someone was "really born as" that you would never ask a
+ straight or cis colleague. The same standard applies.
+
+
+ If you find yourself wondering about surgery, or transition, or someone's identity before you
+ knew them — that's a question for Google, or a community resource, or a book. Not a person.
+ The distinction is this: does this question serve them, or does it serve your curiosity?
+
+
+
3. Their Story Is Theirs
+
+ Being out to you does not mean being out to everyone. Someone trusting you with that
+ information is not permission to share it. Not as gossip. Not as helpful context. Not even as
+ a compliment. Not your story to give.
+
+
+ This matters more in distributed teams than people realise. A Slack message in the wrong
+ channel, a comment on a call with people they don't know, a well-meaning mention in a
+ one-on-one — all of these can out someone without any malicious intent. The rule is simple:
+ unless they told you it's fine to share, it's not.
+
+
+
4. Be the One Who Says Something
+
+ Speak up when queer people aren't in the room. That's exactly when it matters most. The
+ comment that would never get made in front of a queer colleague gets made because everyone
+ assumes no one in the room cares. You can change that.
+
+
+ You don't have to deliver a perfect speech. You don't have to have the right words memorised.
+ "I don't think [person] would love hearing that" is enough. That's the whole thing.
+ This is the biggest gift on this list, and it costs nothing except a moment of choosing to
+ say something instead of saying nothing.
+
+
+
5. Language Is a Welcome Mat
+
+ Job postings, internal docs, emails, Slack messages, all-hands announcements — the language
+ we use by default shapes who feels like they belong. Gender-neutral language costs nothing and
+ includes everyone.
+
+
+ - Use "they" as a default pronoun when you don't know someone's.
+ - Use "partner" when you don't know someone's relationship structure.
+ - Reconsider "guys" as a group term — "everyone," "folks," "team" all work.
+
+
+ None of this requires a policy change or a manager's sign-off. It's a choice, made in the
+ moment, every time you write something.
+
+
+
+
+ Your Action List
+ This is intentionally short. All of it is doable.
+
+ -
+ Right now:
+ Add your pronouns to your Slack profile and Zoom name.
+
+ -
+ This month:
+ If you can donate to an LGBTQ+ charity or mutual aid fund, do it. Every dollar counts, and there's no shortage of organisations doing necessary work right now.
+
+ -
+ This year:
+ When you vote, think about the specific people in your life whose lives are literally on the ballot. Not abstractly. Specifically.
+
+ -
+ Always:
+ When something makes a queer colleague feel smaller — say something. You don't have to get it perfectly right. You just have to show up.
+
+
+
+
+
+ The Close
+
+ That's it. That's the whole talk. And I know it can feel like a lot when you list it out —
+ but most of it comes down to one thing. Treating queer colleagues like their existence is normal
+ and welcome. Because it is.
+
+
+ Every time you use someone's correct name, every time you speak up in a meeting, every time you
+ add your pronouns to a profile — you're giving back energy that person would otherwise spend
+ just trying to exist at work. That energy doesn't disappear. It goes somewhere much more
+ interesting.
+
+
+ You are how we cultivate a remote workspace. Happy Pride. 🏳️🌈🏳️⚧️
+
+
+
+
+
+
+ Back to all talks
+
+
+
+
diff --git a/talks/index.html b/talks/index.html
new file mode 100644
index 0000000..9ca0377
--- /dev/null
+++ b/talks/index.html
@@ -0,0 +1,114 @@
+
+
+
+
+ Voice AI Agent Workshop
+
+
+
+ 20–21 June 2026
+
+
+
+ UC Berkeley Campus, Berkeley, CA
+
+
+
+ Speaker & Sponsor (Berkeley AI Hackathon)
+
+
+
+
+
+ Voice AI is having a moment — and it's more accessible than you might think. In this
+ workshop, we'll build a fully functional voice AI agent from scratch, using real-time
+ speech-to-text, a large language model for reasoning, and text-to-speech to talk back. By
+ the end, you'll have a working agent on your laptop that you can drop straight into a project.
+
+
+
+ Talk to your hackathon project in 40 minutes.
+
+
+ No prior voice AI experience needed. If you've worked with an API before, you're good to go.
+
+
+
+
+ How a Voice Agent Works
+
+ A voice agent is three pieces wired together in a loop. Understanding this loop is the whole
+ conceptual foundation. Everything else is implementation detail.
+
+
+
+ Ear
+ Speech-to-Text
+
+
→
+
+ Brain
+ LLM
+
+
→
+
+ Mouth
+ Text-to-Speech
+
+
→
+
+ The ear captures your audio and transcribes it in real time using
+ speech-to-text. Latency here is everything: if your STT is slow, the whole agent feels
+ sluggish. Deepgram's STT runs at sub-300ms end-to-end, which is fast enough to feel like
+ a real conversation.
+
+
+ The brain receives the transcript and decides what to say. This is your
+ LLM: it reasons over the conversation history and your system prompt, generates a response,
+ and can call any functions you've given it access to. It can look things up, run code,
+ fetch data — anything you wire in.
+
+
+ The mouth takes the LLM's text response and streams it back as audio. Fast
+ streaming matters here too: you want audio to start playing before the full response is
+ generated, or the agent feels like it's thinking too hard.
+
+
+ With Deepgram, all three pieces run over a single WebSocket connection. You're not juggling
+ three separate APIs — it's one socket, one loop, about 80 lines of code to start.
+
+
+
+
+ Design Decisions That Actually Matter
+
+ Most tutorials skip these. They're the difference between an agent that feels like a demo
+ and one that feels like a tool.
+
+
+ Interruption Handling
+
+ In a real conversation, you don't wait for the other person to finish before you start
+ talking. Your agent shouldn't either. When the user starts speaking mid-response, the agent
+ needs to stop its output and listen. Getting this wrong is the fastest way to make an agent
+ feel robotic.
+
+
+ Deepgram's Voice Agent API handles this for you. The WebSocket connection detects voice
+ activity on both ends and manages the interrupt logic. You don't have to implement it
+ yourself — just don't override it.
+
+
+ Conversation State
+
+ Every turn in the conversation needs to be threaded correctly. The LLM needs the full
+ context of what's been said so far — by both sides — to give coherent responses. This
+ means maintaining a message history and passing it in with every LLM call.
+
+
+ Keep your context window in mind. Very long conversations will eventually push older turns
+ out of the context. A simple solution: keep the system prompt, the last N turns, and let
+ the rest roll off. The agent won't remember everything, but the conversation will stay
+ coherent.
+
+
+ Latency
+
+ The number that matters is end-to-end: from when the user stops speaking to when audio
+ starts playing back. Under 500ms feels conversational. Over 1 second feels like a phone
+ call with bad signal.
+
+
+ Three places latency hides: STT processing time, LLM first-token time, and TTS start time.
+ Streaming helps with all three. Start playing audio as soon as the first TTS chunk arrives.
+ Choose an LLM model that prioritises speed over capability for conversational use — a
+ smaller, faster model is often better here than a smarter, slower one.
+
+
+
+
+
+
+ Getting Set Up
+
+
+
Before you start:
+
+ - Node 20+ or Python 3.11+ installed
+ - A code editor
+ - A free Deepgram account — includes $200 credit on sign-up, no credit card needed
+
+
Five steps to a running agent:
+
+ -
+
+ Sign up at
console.deepgram.com.
+ Your account comes with $200 in free credit — more than enough for this workshop and a weekend of hacking.
+
+ -
+
+ Grab your API key from the console. It lives under API Keys in the left sidebar. Copy it somewhere safe.
+
+
+ -
+
+ Clone the starter repo. We'll walk through this together in the workshop, but the pattern is:
+
git clone <starter-repo-url>
+cd <repo-name>
+
+ -
+
+ Add your API key to the environment. Create a
.env file in the project root:
+
DEEPGRAM_API_KEY=your_key_here
+
+ -
+
+ Install dependencies and run it:
+
# Node
+npm install && npm start
+
+# Python
+uv venv && uv pip install -r requirements.txt
+python main.py
+
+
+
+ If it's working, you'll see a connection message in the terminal and the agent will greet
+ you when you speak. If it's not working, come find us at the booth.
+
+
+
+
+
+
+ Make It Yours — Three Modifications
+
+ The starter app works, but it's generic. These three modifications are where the session
+ becomes yours. Each one takes about five to seven minutes. Do them in order, or skip to
+ whichever interests you most.
+
+
+
+
Modification A: Change the Personality
+
+ The agent's system prompt is what defines who it is. Find the system_prompt
+ variable in the starter code — it'll look something like this:
+
+
system_prompt = "You are a helpful assistant."
+
+ Replace it with something more specific. Give the agent a name, a role, a point of view.
+ For example:
+
+
system_prompt = """You are Alf, a friendly AI assistant at a hackathon.
+You give encouragement, suggest project ideas, and answer questions
+about voice AI. You're enthusiastic but concise - people are busy building."""
+
+ Restart the agent and talk to it. The change is immediate. The system prompt is the
+ entire personality — there's no training, no fine-tuning. You just wrote it.
+
+
+
Modification B: Swap the Voice
+
+ Deepgram's TTS has a full voice catalogue. Find the voice configuration in the starter
+ code — it'll be a single string like "aura-asteria-en". Change it to any
+ other voice from the catalogue.
+
+
+ A few to try:
+
+
+ aura-asteria-en — warm, conversationalaura-orion-en — deep, authoritativeaura-luna-en — clear, neutralaura-zeus-en — bold, energetic
+
+ Restart and talk to your agent. It's a one-line change and the agent sounds entirely
+ different. This is the modification that gets the strongest reaction.
+
+
+
Modification C: Add a Custom Function
+
+ This is the one that makes you realise you can hook it to anything. Function calling lets
+ the agent invoke Python (or JS) functions you write, then incorporate the result into its
+ response naturally.
+
+
+ Here's a simple example: a function that returns a random project idea when the agent
+ is asked for inspiration.
+
+
import random
+
+def get_project_idea():
+ ideas = [
+ "A voice-controlled to-do list that reads back your tasks",
+ "An AI study buddy that quizzes you out loud",
+ "A real-time translator that speaks back in the target language",
+ "A voice journalling app that summarises your entries",
+ ]
+ return random.choice(ideas)
+
+ Register this function with the agent and tell the LLM when to use it via the system
+ prompt: "When asked for a project idea, call get_project_idea() and share the result."
+
+
+ Now ask your agent for a project idea. Watch it call the function and weave the result
+ into a natural spoken response. This is the "aha" moment: the agent can call your own
+ code, your own APIs, your own data. The voice interface is just the front door.
+
+
+
+
+ What You Can Build From Here
+
+ Now that you have a working, personalised, function-capable voice agent, here's what that
+ unlocks for your hackathon project:
+
+
+ -
+ Accessibility layer — add voice input and output to any existing
+ interface. Users who can't type or read small text get a completely different experience.
+
+ -
+ In-game NPC — drop the agent into a game as a character that actually
+ talks back. Hook the function calling to your game state so it knows what's happening.
+
+ -
+ Voice-controlled developer tool — talk to your build process, your
+ deploy pipeline, your monitoring dashboard. Voice is an unusually good interface for
+ things you want to do hands-free.
+
+ -
+ Multilingual support — Deepgram's STT handles dozens of languages.
+ The LLM can respond in whatever language the user speaks. Global voice interface, almost for free.
+
+
+
+
+
+ Get Help
+ Building something? Stuck on something? Here's where to find us:
+
+
+ The Deepgram challenge prize this weekend goes to the team that builds the most creative
+ voice-powered experience. Come say hi, show us what you're building, and let us know if you
+ want feedback on your voice integration before judging.
+
+
+
+
+
+
+ Back to all talks
+
+
+
+