A product proposal for the speedrun talent network
Job hunting
sux less.
I used the talent network as a job seeker. These are the problems I hit, and how I’d solve them: first by improving what’s already there, then by building something new.
two working demos inside, plus some memes. you’re welcome ✌You asked: show us how you’d move one of these
Here’s what you’re trying to do. Here’s where each idea lands.
Help job seekers find work they love
less time opening tabs, more time on roles that fit
Grow the network: signups & placements
a reason to join, return, and land the job
Bring great founders to speedrun
better candidates, less screening work
my bet: start with 01. better fits for job seekers are what grow 02 and attract 03.
Improve what exists.
The salary data, filters and personalised ordering are already good. My fixes make them easier to decide with: more of the right detail, explained, in more places.

Six cards called “Founding Engineer.” Which one is backend?
“I keep opening jobs just to find out whether the work, pay and location actually fit me.”
- Same broad title on many cards. The title alone doesn’t say what the work is.
- Salary shows on some cards and not others. Equity never shows.
- “Matches a profile skill” is a great start, but which skill?
Founding EngineerFounding Engineer
Keep the card. Add one row of facts I decide with.
- Role focus + 2–3 key skills: Backend · Python · PostgreSQL.
- Pay with currency and period, and equity on its own line.
- Where I can actually work: “Remote · US only” isn’t “work from anywhere”. Visa if stated.
- A specific match reason, plus the conflict: “Python + API work matches; starts later than you said.”
- Missing data says “Salary not listed” / “Equity not listed”. Never a guess, never “no equity”.
The companies page says “AI Infra.” The job filters have never heard of it.
“I can see which industry a company is in, but I can’t use that industry to find its jobs.”
- /companies groups companies into 13 industries: AI Infra, Frontier AI Labs, Dev Tools & Infra, Fintech…
- The job board’s filters are job functions: Engineering, Hardware & Manufacturing, Operations, Sales… and even inside All Filters there’s no option for the industries /companies uses.
- “Backend at AI Infra companies” means browsing company by company, though the public API already takes an
industryparameter. This may be mostly UI work.


One vocabulary. Jobs filtered through their company.
- Add a company-industry filter using the exact categories from /companies, with live role counts.
- Industry ≠ function ≠ role focus. “Backend at an AI Infra company” uses all three, independently.
- Role focus inside Engineering: Backend, Frontend, Full-stack, Infra, AI/ML.
- Deal-breaker filters: where I can work, salary floor (same currency only), equity listed.
- A “show roles with unknown pay or eligibility” switch, so nothing good silently disappears.
- Clicking “AI Infra” on a company page opens /jobs already filtered.
“Similar roles” is nice. Roles that fit me are better.
“The job page suggests similar titles. I want the roles that fit me, starting with others at this company.”
- Detail pages suggest roles similar to this one, not roles matched to my profile.
- If this company has 20 openings, I can’t see which of the other 19 fit me without leaving.
- There’s no quick way to say “not for me, because of pay / location / focus”.
other roles here that fit me
me
similar titles, anywhereSplit suggestions in two, and explain them.
- “More at this company, for you” right on the job page and tile: the other roles there that fit my profile.
- “Matches your profile” across the network vs. “Similar to this role”, clearly labelled.
- Each with a reason and a conflict, never a mystery percentage.
- One-tap “Not relevant: pay · location · focus” that tunes future suggestions.
- Logged out? You still see “Similar”. Matching is a reason to join, not a wall.
67 open roles. But what do they build?
“I found a company in an industry I care about, but I still don’t know what it builds, who’s building it, or which of its jobs fit me.”
- Collections already show rank, stage, remote policy and role count, with good filters.
- But not a single line about the product, and no count of roles that fit me.
- Company profiles have little founder info and no clear way to reach the team.


One line of context, then the roles that fit.
- A one-sentence product summary on every card, from the official description. No invented traction.
- “3 match you · 67 roles” next to the total count.
- On the profile: For you and All open roles tabs, using the improved cards from 1.1.
- Verified founders with public profiles, and an approved contact route only if the company offers one. No guessed emails.
- Company “remote first” is context; eligibility comes from each role.
83 founding roles. Equity listed on almost none.
“Founding Engineer can mean very different work and pay. I can’t tell from the title whether to open it.”
- The salary tool is genuinely useful: median $198k and a real dilution calculator.
- But the role list below it is title · company · city · pay. Only 34 of 83 roles post a band, and the page itself says essentially none post equity.
- The list can’t be filtered by role focus, remote or equity, or by what fits me.

Reuse the 1.1 cards and 1.2 filters right here.
- Each row gets role focus, a one-line “what you’d build”, remote/eligibility and company stage.
- Equity always has a slot: the % if disclosed, otherwise “Equity not listed”. Then the calculator can run on a real role.
- Filters: focus, remote, salary listed, equity listed, plus “founding roles that fit me”.
- Not a separate feature: same card component, same data.
no new system, just the same cards used on one more page.
Ten worth considering. Then you’re done for the week.
“These are supposed to be for me, but I still sort through different networks and open each role. Give me a small set I can finish.”
- Matching and refresh already exist. But wider market, startup universe and a16z roles sit in one list.
- No way to narrow to speedrun or a16z only without leaving the page.
- Roles start in 2027, but it never asks when I can start.
me, on hour three of reviewing them:


A weekly shortlist you can actually finish.
- Up to 10 roles a week, fewer if fewer fit. Never padded; the live board still has everything.
- Scope right there: speedrun · a16z portfolio · everywhere. Plus start date and preferences.
- Save, Skip, Undo and a list view. Swipe is just a shortcut on phones.
- Save isn’t apply; skip isn’t reject. An optional “why skip?” tunes the future only if I say so.
- “You’re caught up”, saved roles and the next refresh date. An ending, not a feed.
You’re caught up.
Next shortlist on Monday 23 Nov. Here’s what you saved:
Sample roles, not real vacancies.
Build something new.
Your public API and MCP server already do the hard part. These ideas put AI inside the job seeker’s portal, give their agent memory, and take speedrun applications off the Ashby detour.

Describe the job you want. See the filters it picked.
“I know the work, stack and pay I want, but I don’t want to translate that into a dozen filters or set up an external agent.”
- The portal has no chat or AI search today.
- The right roles exist, but reaching them takes a lot of filter clicks.
clicking 12 filters and still not sure what I set
“backend, Python, 120k+, remote from India”Chat in, real filters and real cards out.
- Runs on your existing public search API. No second jobs database.
- Turns my words into visible, editable filters, using only valid values from the live facet list.
- Asks instead of guessing: “120k in USD per year?” “Is Postgres a must?”
- “Remote” ≠ “can work from India.” Unknowns are shown, labelled, never assumed.
- No matches? It says which filter is blocking, and asks before loosening it.
- Guests can use it. “Save as my shortlist preferences?” is an offer, never a gate.
Interview once. Apply without the Ashby detour.
“A form doesn’t show what I know. And every Apply sends me off to Ashby to type the same things in again.”
- Joining the network is a form: no room to show real knowledge or the story behind the work.
- Apply on a speedrun company redirects to Ashby, Greenhouse and the like. New account, re-upload, re-type, every time.
- Founders get a pile of applications with little signal; candidates wait weeks for a silent no.
find the perfect speedrun role
click Apply
redirected to Ashby
type my whole resume in againAI interview at signup → one-click apply → AI screens → founder, or a clear answer.
- Signup ends with a 10-minute AI interview (text or voice) about my real projects. One interview, reused for every application.
- Apply to speedrun companies without leaving: resume + interview summary attached. It can still sync into the founder’s Ashby via its API.
- AI screens each application against the must-haves the founder sets: stack, experience, location, visa.
- Match → handed to the founder with the why and interview highlights. Not a match → a clear status with the reason, in days.
- Kept fair: only the founder’s criteria, never accent, school or polish; human review on request; founders can spot-check rejections.
② apply, without leaving
- AppliedMon
- AI screenTue
- With foundernow
③a match → founder
Asin S. · Founding Engineer- ✓ Python APIs in production
- ✓ real-time voice pipeline
- ✓ remote from India OK
③b not a match → candidate
Not moving forwardThis role needs 3+ years of Go in production; your interview didn’t cover Go.
Request a human review →Connecting an agent shouldn’t need a config file.
“I want to connect my agent without hunting for the right config file or pasting a JSON block.”
- Claude Code already has one command, and Claude web has a custom-connector URL. Both great.
- Other local clients mean finding and hand-editing config files.
one does not simply
find the right config fileClient tabs, a copy button, and an optional guided installer.
- Keep Claude Code’s one command and the Claude connector URL front and centre.
- For other local clients, an installer that asks which client, previews the exact change and asks before writing.
- Backs up the config, is safe to re-run, can uninstall, and checks the connection at the end.
- Installing never collects a password or grants permission to apply.
small, and only worth building if setup is where people drop off.
Let my agent do the searching. I decide where to apply.
“I already live in Claude and ChatGPT. Let them search with my real preferences and send me a short list.”
- Your MCP server already searches live roles with the board’s filters. No auth, no keys.
- But my agent forgets me: no saved preferences or roles shared with the site.
- Nothing comes to me: no weekly shortlist unless I ask.
- Applying is manual, and should stay my decision.
my agent applies to jobs for me
after I approve each one, right?
…
right?Connect once, get a weekly shortlist, apply after a review.
- Sign in with scoped permissions (OAuth, never a password in chat): read approved profile fields, search, save.
- Same preferences and saved roles on the site and in the agent. One shortlist, from 1.6.
- Optional weekly digest sent by the server, not a chat left open. Job info only in the email.
- Prepare, then I approve: drafts use only facts I’ve approved; then one-click apply for speedrun companies (2.2), or the employer’s form elsewhere.
- Job listings are untrusted text: they can’t change permissions or trigger anything.
Claude wants access to:
- Read approved profile fieldsskills, focus, location · not contact info
- Search & save to your shortlistsame list as Roles for You
- Weekly digest by emailverified address · opt-in
- Submit applicationsoff by default · approved per job
- Role
- Founding Eng · Ledgerline
- Resume
- asin-backend-v3.pdf
- Why you?
- from your approved facts
- Work auth
- you tell us, never guessed
Let’s build it together.
Fix what’s there first, then build what’s new on the same cards, filters and matching. Less time opening tabs, more time on roles that fit.

Independent concept by a candidate.