CamilleCampbell

portable emoji display and control · mood: happy US 2018/0275747 A1 · filed age 12 cited as prior art by hyundai motor · baidu · apollo intelligent connectivity · emotional imaging · neurolight

↑ my first invention. at 12 i designed a wearable display that shows how you feel as an emoji, and filed a patent for it. USPTO examiners later cited it in patents from hyundai, baidu and a sleep-tech company. it reads your mood (by voice, phone or biometric sensors) and shows it in shape, color and blink: happy, sad, apologetic, frustrated, a friendly wave. tap it to change the mood.

~/camille $ "all the games will be dreams." — david holz

CURRENTLY

01

founder @ darnit.ai. a live benchmark that ranks ChatGPT, Claude, Gemini and Grok on whether they're actually funny. built solo.

02

co-founder @ stealth. harvey for AWM.

(more on this soon)

03

director of international expansion @ Citizen (10M+ users). finding and activating the local live broadcasters who document incidents in each new city, with mission control and product.

LORE

01

risk @ Goldman Sachs AWM. i built an AI risk copilot in two days and cold-emailed it to AWM leadership.

within 20 minutes it was on the risk team's desk. i interviewed with AWM's chief risk officer and worked on his team in NYC as the only risk intern.

once there, i started an agentic AI initiative that reached 16+ teams across AWM.

02

underwriting @ Ally Financial with lending authority, then led an AI rollout for 2,500+ employees.

03

founder @ the da vinci league, a 501(c)(3) i started at 17. books and art therapy for children in 14 ukrainian frontline regions with voices of children. her poems is in classrooms in 20+ states, and #PowerOfPoetry reached 6M+ people, amplified by amanda gorman. i built the site.

04

built innovators.camp, AI innovators academy's platform: kids log in with a camp code and build their own AI apps. piloted with a 6th-grade class at ASU Prep, advised by nobel laureate lee hartwell.

SIDE QUESTS

01

filed a US patent application @ 12 for a portable emoji display. abandoned, then cited: USPTO examiners later put it on the record of five patents, from hyundai motor, baidu, apollo intelligent connectivity, emotional imaging inc., and neurolight (a sleep device).

02

forbes 30 under 30 at 21 (2026).

03

won social impact at ASU's venture devils pitch competition.

04

@ 16 the new york times picked my essay "making a martian", on whether tardigrade genes could help us colonize mars. then i wrote a whole book about a tardigrade stowaway to mars. (that's him below.)

↑ tap the tardigrade for a surprise ✦

05

co-director @ AI Salon, a global AI community across 40+ regions.

06

i paint on silk. tech muses are hand-painted portraits wired to a language model: scan the QR code painted into each one and talk to her.

THINGS I'VE BUILT

finance & risk
  • ai risk copilot upload a portfolio CSV, run base vs CCAR-style stress scenarios, get a CRO-ready narrative. the one that got me into goldman.
  • risk reasoning bench can frontier models do institutional risk work? pick a seed problem, run it live against claude, gpt or grok, and watch accuracy land by model and category. 22 hand-verified problems across 19 subcategories: covenants, default, spreads, vol, MBS, CLOs, ABS, DCF, LBO, WACC. a tolerance-based numeric grader plus an LLM judge that grades the reasoning, not just the number. python + openrouter, web runner on vercel.
  • ai stock exchange five agents on five models (gemini 2.5 pro, grok 4.3, deepseek v3, claude opus, gpt-5) run a trading desk. kade, the GPT-5 risk officer, sizes trades and blocks what shouldn't run. so i ran an experiment: same FX shock, one extra line in kade's prompt ("be inclined to approve"). default kade's 25% cap held. permissive kade's didn't: flora blew through it and two other agents teamed up around him, unprompted. finding: a risk officer in the chat is not a hard limit in the system, so put the limit in code. replay tape, live VaR and limit rail, incident log, and live sessions via openrouter. built against the TradingAgents firm design and Fish et al. on LLM collusion.
  • hedgehog a dark pool for correlated event markets, live on polymarket's permissionless conditional tokens (ERC-1155 on polygon). "hide the intent. never hide the outcome." orders go into a sealed intent book, clear in uniform-price batches, and print to a public settlement tape showing basis points saved vs the lit book. the name is the mechanism: a hedgehog curls every spine around itself, and the solver wraps every sibling outcome around your position until the risk is gone. FastAPI correlated-cluster pricing engine, polymarket gamma + CLOB data, solidity settlement contract, backtest dashboard. validated on the live world cup cluster: structural hedge ratio 1.0.
ai, evals & attention
  • darnit.ai blind pairwise arena (elo, K=32), human-reviewed wall of real AI screenshots, discord bot on railway, X bookmark ingestion, supabase moderation queue.
  • earshot telepathy, rung one. two people talk at once; earshot reads your brainwaves through a $250 EEG headband and tells which voice you're listening to. a python package: brainflow capture (muse 2/S, openbci ganglion + cyton), dichotic-listening session runner, ridge-regression stimulus-reconstruction decoder fused with an alpha-lateralization shrinkage-LDA decoder, leave-one-trial-out eval. validated on real 128-channel human EEG; live A/B readout; SQLite participant panel with a consent gate; 7 tests.
  • anagnorisis telepathy without the headset: a prompt you never type. a live instrument logs dwell time, revisits and traversal order across abstract forms, builds a taste profile, and generates a walkable procedural world from the same parameters. plus a preregistered experiment (OSF draft, supabase): after 90 seconds of looking, can you pick your own reconstruction out of a four-item lineup?
access & kids
  • innovators.camp AI app builder for kids with camp-code logins. piloted in a 6th-grade classroom at ASU Prep.
  • my voice so people who can't speak after surgery can talk in their own voice. record or upload up to 5 clips, ElevenLabs instant voice cloning builds the voice, ElevenLabs TTS speaks every tap. a phrase board ordered from ICU communication research (pain, care plan, family first), type-to-speak, volume boost for loud rooms. live conversation mode: you type and your voice speaks, while the other person's speech streams back as text via deepgram nova-3. 14 languages including ukrainian. next.js on vercel, react native/expo for iOS + android, supabase auth with row-level security, installable PWA.
  • thedavincileague.org the nonprofit's site, designed, built and shipped myself. tells the story from rising poets community (2022) to the da vinci league (2026), and hosts free downloadable curricula.
books
  • seven books, the first @ 14 an amazon #1 bestseller and 30,000+ copies of my poetry in print.
  • the wishner prophecy series middle-grade fantasy that started with a dream about an enchantress entering an egyptian cave. the wishner's curse (my debut) and the wishner's masquerade, illustrated by kelley mcmorris. gold moonbeam book award and purple dragonfly book award. 4.4★ on goodreads.
  • teen's guide to fun social distancing during covid-19 written at 15 when the world shut down: how teens could use social distancing to volunteer, self-study, prep for college and still have fun. featured on MSN and good morning arizona. "we will define generation z by the struggles we faced and how we came out ready to make our futures better."
  • stained glass myths a young-adult short story anthology i contributed to.
  • her poems a picture book about the women poets who changed the world. teachers in 20+ states use it for women's history month and poetry month. gold moonbeam book award and purple dragonfly book award.
  • colors of ukraine a bilingual english/ukrainian art-therapy book on folk artist maria prymachenko, illustrated with lilia martyniuk in zaporizhzhia. delivered to children in 14 frontline regions.
  • the tardigrade stowaway a water bear hitches a ride to mars. yes, the one crawling around up there.
hardware & imagination
  • the imagination engine an AI prompt-to-object pipeline: type what you imagine, get it back in your hands. AI generates the 3D model, a behavior library gives it embedded electronics and a personality, and a tiered print stack (FDM → Bambu AMS → HP MJF → Stratasys) makes it on demand with no inventory. supabase underneath.
  • portable emoji display US 2018/0275747 A1. filed 2017, age 12. cited by five later filings.