ADIN

Scouts and founders submit startups. The software gathers material on each one and drafts an investment report. The team reviews it and votes, and funded companies move into the portfolio. One place to source, research, decide, and track.

I built the public site and the ways into the product from it: joining as a scout, submitting a deal, and the onboarding that follows. The rest of this page follows the same flow the product does.

Over a thousand people find companies for ADIN. Most aren't investors and open the product occasionally, so anything asked of a scout has to be quick to finish and hard to get wrong.

I own deal submission end to end: the multi-step form, and the pipeline that turns a submission into a durable record, ingests its files, and starts a report. Most of the engineering is invisible: large uploads, submission locking, duplicate prevention, retries, so a flaky connection can't leave two half-finished deals. Referral attribution starts here too, and the rewards in section 05 are built on it.

What an investor needs on an early-stage company is scattered across the submission, the deck, the site, market and competitor data, technical detail, and press. Reading all of it for 1,300 submissions isn't something a small team can do.

So research runs as a pipeline. Sources are collected, relevant material is retrieved, and agents each work one angle: team, market and competitors, product and technology, coverage, and the diligence questions the fund asks of everyone. Their outputs are structured rather than free text, so a report can be assembled, cited, and regenerated one section at a time.

I work on that platform with the other founding engineers and own the seam between it and the product: long-running async jobs, status tracking, and recovery when a step fails. Python for the pipeline, PostgreSQL and cloud storage for the reports, TypeScript and Next.js on GCP for the product.

Generated research is only worth something if people can work with it. Reports pull every agent's analysis into one surface, and I lead the frontend architecture and interaction design for them: how an investor moves between analyses, checks the evidence behind a claim, and works with generated text.

The page is dense, so hierarchy does the work. Summary and recommendations first, each analysis with its sources attached, the full report and comments at the end, and voting visible from anywhere on the page.

AI supports the call; people make it. The team votes, edits any section, regenerates one when the material changes, and follows the revision history. I built those editing and regeneration flows, plus password-protected sharing so a report can go to a co-investor without opening up the platform.

A network keeps producing only if contributing is worth the time. Members earn for what helps the fund: submitting deals, referrals, completing a profile, verified social activity.

I built points as a ledger. Every award is a transaction against a configurable earn event, with eligibility rules and duplicate protection, so balances can be recomputed and audited rather than trusted. Members get tasks, balances, history, and leaderboards; internally there are tools for campaigns, activity review, and adjustments. Awards run as background jobs, so a burst of activity never slows the product down.

Once a company is an investment the questions change: what we put in, on what terms, how it's doing, and what the fund looks like as a whole.

I work on portfolio records, investment data and monitoring, fund information, and the internal admin tooling behind them, along with the content tooling the team publishes with. A little of it is member-facing, like recent investments on the public site. Most is internal, used by a handful of people every day, and built to change as fast as the fund's process does.

ADIN is live and still being built. The bet behind it is that a fund can run on software: sourcing, diligence, and the decision itself in one system, with a network of people feeding it. That is what we are building, and a lot of it is still ahead.