I'm the co-founder and COO of Bluedot,
a Y Combinator (W23) company that built software and payments infrastructure for EV fleets,
acquired in 2026. We raised $5.5M+ and served customers like Rivian, Hertz, Genentech, Duke Energy,
the City of San Francisco, and the City of Seattle.
At Bluedot, the market didn't come with clean budget lines, obvious buyers, or established
language. We had to make it legible first then build products fleets could actually adopt.
That meant turning fragmented charging, reimbursement, and payments workflows into software,
and building the go-to-market motion around it from scratch. I led product discovery and
enterprise sales across complex buying groups: operations, finance, procurement, drivers,
and shaped both what we built and how we sold it.
I grew up in Turkey and moved to the US after college to start a company. I studied
psychology and business at
Boğaziçi University
and later spent time at
Nova School of Business and Economics.
Before Bluedot, I built in-car applications for Mercedes, Audi, and Renault,
launched a textbook-sharing platform, and co-created
TOKTUT, a nonprofit
that helped distribute more than 1.5 million meals during the pandemic. More recently, I built
DECISION.md,
an open-source project about encoding judgment into AI agents.
I'm interested in decision-making, poker, mental health, and robotics. Most of all,
I'm interested in how AI and technology get implemented in the real world and the physical
economy, and the impact they have on how we live and work.
projects
an open-source framework for encoding decision-making judgment into AI agents. adds the missing layer between agent identity and capabilities: how an agent should handle tradeoffs, calibrate confidence, and navigate risk.
where it sits in the agent stack:
SOUL.md
identity: who the agent is, its values and personality
SKILL.md
capabilities: what the agent can do, its tools and knowledge
AGENTS.md
coordination: how agents delegate and work together
DECISION.md
judgment: how the agent handles tradeoffs, risk, and uncertainty
the missing layer
try it: answer five tradeoff questions, get a mini DECISION.md snippet
software and payments infrastructure for electric fleet operations. charging access, reimbursement workflows, vehicle analytics. Y Combinator W23, acquired in 2026.
infrastructure
fleet systems
yc w23
acquired
a dish-first san francisco restaurant guide. pick a dish, see the spots ranked on two signals read from thousands of google reviews: how often the dish gets mentioned, and how positively people describe it. 300+ spots, 60+ dishes, a michelin ingredient-sourcing map, and a data-driven blog. companion newsletter: sf gems.
consumer
discovery
san francisco
Selo
a private social app for the things you've actually kept. six categories that mirror how people organize their lives. anti-consumerism is enforced in the data model: you can only save things to try from a friend's shelf or a letter. currently in TestFlight beta.
react native
convex
ios
social
an iOS podcast note-taking app. capture voice notes while listening with automatic timestamping. a book club for podcasts: small groups discuss episodes together with synced listening and local voice transcription.
ios
podcasts
social
a research-grounded, pre-engagement compatibility and readiness assessment for couples. individual questionnaires across ten relationship dimensions plus six guided scenario conversations produce a couple profile, a dimension gap map, a conflict portrait, and a three-move action plan. grounded in published marriage research (Karney & Bradbury, Gottman & Levenson, Olson & Fowers, Huston's PAIR project).
relationships
research
assessment
one object per year, tracing American history from 1776 to 2026 through the physical things it made and kept, from the cotton gin to the iPhone. told with equal weight given to triumph and hardship, not just celebration. built for July 4, 2026.
history
archive
americana
a curated collection of traditional dishes from around the world built on surprising ingredient combinations that shouldn't work on paper but are genuinely beloved in their place of origin. no shock value: every dish is verified against published sources.
food
culture
archive
a reinforcement learning project that trains a small bipedal robot to autonomously grasp and deliver a book across an apartment. the trained policy (PPO on MuJoCo Warp) hit 100% delivery success over 32 rollouts and is exported as a deployable ONNX model.
reinforcement learning
robotics
pytorch
a local-first reading room. bring any article, mark it up as you read, and think it through with Claude. every highlight asks you why you marked it, notes live in the margin next to the passage, and the discussion knows the whole piece. everything stays on your machine.
reading
ai
local-first
a weekly letter from your AI, about you. a private local note grounded in your recent Claude Code or Codex conversations: what the agent actually completed, what changed in its model of you, and one question worth sitting with. silent by default: if nothing meaningful happened, no letter. open source.
open source
ai agents
local-first