I wanted to share my first AI agent build that I have done a couple months back! I designed and built Cosmo, a multimodal voice agent combining n8n for orchestration, Groq/Llama for response routing, ElevenLabs for natural text-to-speech, the Web Speech API for voice input, and Supabase for persistent memory, wrapped in a custom retro console-style interface. The architecture has changed under the hood and it has been so fun to use my own personal Jarvis! https://www.instagram.com/p/DZfHudqx1jf/
6
274
AWS "Agents for Humans" Hackathon ā Nonprofit Ops Agent
1
129
Signl Health is a mental health tracker built for documentation, not just reflection. It logs daily mood and patterns and turns them into clear evidence you can bring to a doctor ā instead of trying to recall months of how you've been feeling in a short appointment. It also supports tracking for more than one person, so you can log for yourself and separately for a child or dependent. Built entirely in Flowstep, wired through their MCP server via Claude.
#FlowstepChallenge
https://app.flowstep.ai/file?activeFileId=6a5ceee7-fa33-4091-8960-340fbb5e4da4
1
4
242
I built My Little Bot ā a playful custom AI bot platform where users can create, personalize, and chat with their own bots. š¤āØ
What started as a fun idea turned into a working mini SaaS-style app with:
⢠Custom bot creation
⢠Bot personalities and instructions
⢠Saved conversations
⢠Voice options
⢠Premium upgrade features
⢠Bot templates
⢠Authentication
⢠Admin tools
⢠Stripe subscription flow
⢠Netlify deployment
The goal was to make AI feel more personal, creative, and approachable ā something users can shape around their own routines, ideas, projects, or support needs.
People can create bots like:
study buddies, brainstorming partners, creative assistants, accountability bots, character-style companions, or task-specific helpers.
This project brings together a lot of the work I love doing: AI product design, user experience, playful interfaces, prompt systems, automation, and creative technology.
Iām currently inviting people to test it and share feedback on what feels intuitive, what breaks, and what could make it more useful.
Try it here: https://mylittlebot.netlify.app
Built with Fable 5, Supabase, Stripe, Anthropic, and Netlify.
One of the more interesting challenges in payroll is making financial obligations feel immediately understandable. Organization admins donāt just need to know whatās dueāthey need confidence that payroll is covered, visibility into whatās driving the total, and a clear next step when it isnāt. This concept explores that idea by turning payment obligations into a single, actionable experience: surfacing funding status at a glance, breaking down where costs come from, and providing AI-powered answers to the questions admins are most likely to ask. The goal wasnāt to add more data, but to reduce uncertainty around one of the most critical moments in the payroll lifecycle.
1
224
I wanted to see how far I could push simplicity. Instead of asking merchants to interpret dozens of charts and KPIs, the experience starts with a single health score that quickly communicates the current state of the business. From there, supporting metrics provide just enough context before guiding attention to the one issue that matters most. The goal was to reduce cognitive load without sacrificing confidence in the data.
2
267
Iāve been exploring how interfaces can tell stories instead of simply displaying information. This concept uses a particle-based reconstruction engine to transform fragmented sensor data into an evolving visualization of something ancient, immense, and still largely unknown. Over the next few days, Iāll be sharing a fully interactive prototype that brings the entire system to lifeāfrom AI interpretations to real-time analysis and evolving reconstructions.
Form Field Rearrangement A/B Experiment
Problem:Ā Analytics showed a heavy drop-off on the final step where users were asked for name, phone, and email all at once, it felt like handing over everything in a single ask.
Action:Ā Restructured the form by moving the email field to an earlier step, separating it from the phone field so the contact ask felt more progressive and less invasive.
Result:Ā The repositioned email field reduced friction at the final step and lifted overall conversions by 3.48%.
0
71
Video Thumbnail Design A/B/C Experiment
Problem:Ā After launching the Medicare Thank You Video, watch rates were lower than expected, the default thumbnail wasn't compelling enough to pull users into the video.
Action:Ā Designed and tested 3 new high-contrast thumbnail variations against the original, optimizing for visual interrupt and click-to-play intent.
Result:Ā All 3 variations beat the control on lead-to-policy, with the winning thumbnail lifting LtP from 4.3% to 5.9%.
1
100
Adding Thank You Video After Form Submission A/B Experiment
Problem:Ā After a Thank You Video boosted performance in our Medicare flow, we wanted to validate the same play in our Life Insurance funnel to drive higher lead-to-policy conversion.
Action:Ā Tested two Thank You Video variations on the LifeCompared confirmation page, one AI-generated and one featuring our director on camera, measuring downstream lead-to-policy.
Result:Ā The director-led video won, lifting lead-to-policy from 5.8% to 6.9% and confirming that authentic, human-led video outperformed the AI alternative.