Aadi Shankar
I study neuroscience and business at UC Berkeley. I research how the brain ages, and I build products that make complex systems feel simple.
This brain is the site map. Pick a region.
Takes
Quick reads on AI, finance and M&A, neuroscience, and product and design, written while the news is still news.
One Anthropic researcher puts the odds that AI kills us all within a decade above 10%. I went through the research to see how software we built could ever stop taking orders.
Read the take
Psychedelics went from banned to a White House priority in a few years. I dug into what they do in the brain, what the trials actually show, what users report, and who stands to make money, to answer one question: is the hype real?
13 min read
I wanted to understand why AI-generated video keeps winning the scroll, so I looked at where it's made, what people say about it, how its scripts are built and what neuroscience says about why we keep watching.
7 min read
In the drafts folder
A note on why the best decision tools should not promise certainty, but should help people understand the shape of what they might regret.
Dense research is not only a comprehension problem. It is a navigation problem, a visual hierarchy problem, and increasingly an AI product problem.
The college social graph is not the same as the internet social graph. It is temporal, physical, and full of half-formed intent.
Markets are interfaces for fear, status, patience, prediction, and time. Better finance tools should make those forces legible.
Now
Two products in progress, and room for one more.
A pixel-art M&A tycoon game. The iPhone and iPad beta is in TestFlight, and a free web demo is live.
See the gameOne shared memory for every AI assistant you use, so Claude, ChatGPT, Codex and Cursor pick up where the others left off.
How it worksI am open to building more ideas, especially where neuroscience, AI and finance overlap. Bring a problem worth solving.
Reach out on LinkedIn(opens in a new tab)An M&A tycoon game
Start a boutique bank in a garage. Win mandates, outbid rivals, survive raids, and become a Wall Street Legend.
You have 52 weeks to grow from half a floor to a penthouse tower, against three rival banks chasing the same mandates. Every pixel is drawn in code.
01 / 03
Read the founder's tell, play the cards they love, and beat three rival banks to the mandate.
02 / 03
Line up buyers, push the top bidder, and squeeze every last dollar out of the deal.
03 / 03
Hire a team, survive rival raids, and move up from a garage boutique to a penthouse tower.
HTML, canvas and vanilla JavaScriptabout 24,000 lines, no dependenciesCapacitor 8 for iPhone and iPada test harness plays full seasons to check balance
The memory and teamwork layer for AI assistants
Most AI work dies in a chat window. ModelSync gives every assistant you use the same memory, keeps the cost of picking work back up low, and lets people and models split real projects: who plans, who builds, who reviews.
Open ModelSync (sign-in)(opens in a new tab)How a Relay run moves
Nothing ships until a human says so.
Claude, ChatGPT, Codex, Gemini CLI and Cursor write facts, decisions and handoffs to one project record, so any of them can pick up where another stopped.
Assistants read only what changed since their last visit, a short briefing instead of a replayed transcript, so every session starts on fewer tokens and work goes to whoever has capacity left.
Shared workspaces tied to a GitHub repository. Members share only the chats they choose, and implementation and review are routed by each person's available capacity.
Claude and Codex plan the same objective on their own, a deterministic merge decides what survives, each builds in an isolated worktree, and they review each other's work.
Product design
Each one starts with a behavior I noticed and ends with a product plan. Pick a screen to see it.
Concept and roadmap
Your memories were never just a single frame
ReLive uses generative AI to rebuild the moments around a photo: the seconds before and after, the conversation, the place in 3D, and how one memory connects to the rest. The roadmap takes it outdoors, turning real cities into explorable worlds with AI-guided journeys.
Working Mac app
Privacy for conversations in public spaces
A desktop privacy layer that hides older messages by default and leaves only the last one, two or three visible. Hover over a hidden message to reveal it for a moment with an invisible-ink effect. No message content is ever stored.
Data product concept
See which students thrive, and why
Outcome-informed admissions intelligence: colleges look back at real student outcomes and compare them with applicant profiles, with fairness monitoring and bias guardrails built in. Advisory only; it supports decisions and never automates them.
Learning platform concept
Understand scientific papers without losing the science
An AI learning workspace that turns dense papers into reading paths, interactive figures, mechanism animations and claim-to-evidence maps, so a reader can follow the argument without losing the rigor.
About
I am an undergraduate at UC Berkeley studying neuroscience and business. I spend my weeks between a brain imaging lab, private equity work and the products I build at night. The thread through all of it: finding the hidden structure behind how people decide, and turning it into something useful.
$50K–$100K
potential savings in technology costs
Private equity
Financial modeling and analysis as a private equity analyst at Sansara Group, and now at HCAP Partners in San Diego. I also redesigned a CRM for a private equity firm with all the functionality of DealCloud, bringing deal sourcing, pipeline and portfolio tracking into one system and potentially cutting up to $50K to $100K in technology costs.
700+
participants in the HEAD study
Computational neuroscience
Research Intern in the Jagust and Baker labs at Berkeley Lab. I work on PET quantification: computational methods that turn brain scans into measurements you can trust across people, scanners and years. The work feeds the HEAD study, was first presented at Human Amyloid Imaging 2026, and is heading toward a paper. Supported by a $10,000 LSBE Research Scholarship.
2000+
counselor responses trained and scored
AI for mental health
At the Vandrevala Foundation I built and designed an AI-powered co-pilot on modified LLMs that trains crisis counselors on India's largest mental health chat and call line, which serves 600,000+ patients across underserved states. It earned commendation from the leadership team and the counselors, and contributed to a Stanford University research project.
Product design
I design products end to end, from the behavior that sparks them to the screens and the roadmap. Beyond Buyside and ModelSync: ReLive, which rebuilds the moments around a photo; Veil, a privacy layer for messages; Trajectory, admissions intelligence; and Signal, interactive scientific papers.
See the designs →Research
PET scans use a tracer to show where proteins such as tau build up in the brain. The hard part is turning those images into numbers you can compare between people, scanners and visits. That quantification is what I work on, with computational methods for measuring the signal cleanly.
Much of it feeds the HEAD study, which scans more than 700 older adults across nine sites with several tau PET tracers side by side, so results from different tracers can finally be compared on one scale. I am in the process of publishing a paper that contributes to this work. The research was first presented at the Human Amyloid Imaging Conference 2026.
Hover a role to open it.
Dec 2024 - Dec 2025
Vollie
Part-time · 1 yr 1 mo · Berkeley, California · On-site
A volunteering platform piloted with the Berkeley community, matching students with local organizations and campus initiatives. Over 100 users joined and we ran 4 community events before the pilot wrapped in fall 2025.
Supported and funded by the ASUC Social Impact Incubator.
vollie.ai ↗ (opens in a new tab)Founding
May 2025 - Sep 2025
Sansara Group
Part-time · 5 mos · Hybrid
Private equity
May 2025 - Sep 2025
Vandrevala Foundation
5 mos
I built and designed an AI-powered co-pilot using modified LLMs, training and scoring 2,000+ mental health counselor responses and improving crisis message quality across India's largest mental health chat and call line, which serves 600,000+ patients. The tool trains crisis counselors who support people in underserved Indian states without access to healthcare providers, and it earned commendation from the organization's leadership team and its counselors.
Contributed to a Stanford University research project.
AI
NowSep 2025 - Present
Berkeley Lab
Part-time · 1 yr 3 mos · Berkeley, California · On-site
Computational neuroscience in the Jagust and Baker labs: turning brain PET scans into reliable measurements for studies of aging and Alzheimer's disease.
Recipient of the $10,000 LSBE Research Scholarship for laboratory work.
lbl.gov ↗ (opens in a new tab)Research
Sep 2025 - Dec 2025
Siemens Healthineers
Full-time · 4 mos · San Francisco Bay Area · Remote
siemens-healthineers.com ↗ (opens in a new tab)Consulting
NowJun 2026 - Present
HCAP Partners
Internship · 6 mos · San Diego, California · On-site
hcap.com ↗ (opens in a new tab)Private equity
Based in Berkeley / New York
Milestones
2026 / Product
My pixel-art M&A tycoon game: playable on the web, in beta on iPhone and iPad, launching soon.
2026 / Research
My PET quantification work for the HEAD study, first presented at HAI 2026. Paper in progress.
2026 / Product
One shared memory for every AI assistant, live as a hosted beta.
2026 / Private equity
All the functionality of DealCloud in one system, potentially cutting up to $50K to $100K in technology costs.
I am always up for conversations about neuroscience, AI products, private equity and finance tools, and ideas with a strong point of view.
Product decks and deeper case studies on request.
