diff --git a/The todo list.md b/The todo list.md new file mode 100644 index 0000000..3dca92b --- /dev/null +++ b/The todo list.md @@ -0,0 +1,287 @@ +Roles + +https://elevenlabs.io/careers/6b84bd1f-05cb-4b54-b0db-ff64c84217ae/deployment-strategist-lead-nordics + +https://jobs.ashbyhq.com/cohere/2d256112-b336-4539-8133-a0bf7f6698f0?locationId=f15fd0a7-62c8-4e26-a06d-f396a4e21ffb + +https://jobs.ashbyhq.com/mistral.ai/2c1ff089-db74-467c-b17d-608713e43699?departmentId=cdcd26c0-0c1e-4aff-91f7-5f6306b726b3 + +https://jobs.ashbyhq.com/Sierra/44120ba5-4cf7-48a1-86af-7981c41c179e + +https://jobs.ashbyhq.com/black-forest-labs/bf4fa9be-7a1e-4b1e-b6c7-1f399e73f2e3 + +look at byfounders + + + +### 1. Frontier labs and AI companies with forward deployed teams (mostly no Danish office) + +| Company | Why it fits you | Nearest office | +| -------------------------------------- | ------------------------------------------------------------ | ------------------------------------- | +| Anthropic | Applied AI and FDE teams growing fast in Europe | London, Dublin, Zurich, Paris, Munich | +| OpenAI | Forward deployed engineering and solutions teams | London, Dublin, Paris, Munich | +| Mistral | Enterprise deployments for European sovereignty | Paris, London, Munich, Amsterdam | +| Google DeepMind | Research leadership | London, Zurich | +| Cohere | Almost entirely enterprise and private deployment, so heavy on FDE | London, Paris | +| Palantir | Created the FDE role; Danish police run its software (POL-INTEL) | London and other EU offices | +| Scale AI | Its public-sector and enterprise work is FDE-style | London | +| Hugging Face | Open-source, remote-friendly | Paris / remote | +| ElevenLabs | Hires many forward deployed and solutions engineers | London / remote in Europe | +| Harvey, Sierra, Glean, Writer, Decagon | AI apps sold to enterprises, each with FDE teams | London | +| Poolside, H Company | Frontier labs based in Paris | Paris | +| Helsing | Defence AI, deployment-heavy | Munich, London, Paris | +| Aleph Alpha, DeepL, Black Forest Labs | The German AI companies | Heidelberg, Cologne, Freiburg | +| Wayve, Synthesia | UK AI scale-ups | London | +| Databricks, Snowflake, NVIDIA | Platform vendors with solutions and field AI roles. NVIDIA works closely with Denmark through the Gefion supercomputer | Nordic sales offices | + +### 2. Denmark + +**Applied research and public-interest organisations (the Alexandra Institute's peers)** + +- **Alexandra Institute**: applied AI research in Aarhus and Copenhagen. +- **Danish Technological Institute (Teknologisk Institut)**: large robotics and AI centre in Odense. +- **DCAI (Danish Centre for AI Innovation)**: runs the Gefion supercomputer and needs people who help companies use it. +- **Pioneer Centre for AI (P1)**, **DTU Compute**, **Novo Nordisk Foundation**: research leadership roles. + +**AI companies** + +- **Corti**: medical speech AI in Copenhagen, with strong research. +- **2021.AI**: AI governance. +- **Veo**: computer vision for sports video. +- **Milestone Systems**: video AI in Brøndby. +- **Evaxion**: AI-driven biotech. +- **Dixa**: AI for customer service. + +**Large employers with serious AI work** + +- **Novo Nordisk**, **Novonesis**, **Maersk**, **LEGO** +- **Microsoft Development Center Copenhagen** in Lyngby +- **Zendesk**, whose Copenhagen R&D office builds its AI agents +- **Unity**, **Trustpilot**, **Pleo** + +**Consultancies with FDE-like delivery roles** + +- **Netcompany**, **Implement Consulting Group**, **BCG X**, **McKinsey QuantumBlack**. All four are in Copenhagen. + +### 3. Robotics in Odense (and the rest of Denmark) + +- **Start with the Odense Robotics cluster.** It runs its own job board covering about 150 member companies, so it's the most useful single board to watch. +- **Universal Robots** and **MiR**: both owned by Teradyne. UR has an AI Accelerator built with NVIDIA, which is the most AI-heavy work in Odense. +- **OnRobot**: grippers and tooling for robot arms. +- **Blue Ocean Robotics**: UVD Robots and GoBe. +- **Nordbo Robotics**: robotics software. +- **Rope Robotics**: robots that repair wind turbine blades. +- **Quadsat**: drone-based antenna testing. +- **SDU**: the robotics group and the Mærsk Mc-Kinney Møller Institute. +- **Outside Odense**: **Capra Robotics** (Aarhus), **Life Science Robotics** (Aalborg), **Enabled Robotics** (Copenhagen), **Sky-Watch** (drones). + +### 4. Sweden, with Skåne first + +**Skåne (you can commute)** + +- **Ericsson Research**, **Sony**, **Axis Communications** (video AI), **Arm** and **Qualcomm**, all in Lund. +- **Ingka Group Digital** (IKEA), in Malmö and Copenhagen. +- **Verisure**, in Malmö. + +**Stockholm and Gothenburg** + +- **Lovable**: the fastest-growing AI company in Europe. +- **Legora**: legal AI, hiring heavily in solutions roles. +- **Sana**: now part of Workday. +- **Klarna**: AI-first. +- **Spotify**, **King**, **Epidemic Sound**. +- **Kognic** and **Einride**: autonomous driving. +- **Modulai**: ML consultancy. +- **AI Sweden** and **RISE**: public applied research. RISE is Sweden's counterpart to the Alexandra Institute. + +**Elsewhere in the Nordics** + +- **AMD Silo AI** (Finland, acquired by AMD), **Cognite** (Norway), **Oura** (Finland). + +### Where to put your effort + +- **FDE lead roles:** Anthropic, OpenAI, Mistral, Cohere, ElevenLabs, Legora and Lovable. All of them hire in London or Stockholm and many accept remote work in Europe. +- **AI research leadership in Denmark:** Corti, DCAI, the Alexandra Institute, the Novo Nordisk ecosystem and Universal Robots. +- **Robotics:** the Odense Robotics job board plus UR and MiR. + +## Frontend run through + + + +### More sources: + +- Look at more sources (**76 companies are read by hand**, on the Watchlist with a careers link opened on 2026-09-29: 35 `unread`, 10 `robots`, 9 Teamtailor ones without a career host, 8 `group`, 7 `blocked`, 4 `gone`, 3 `key`. The remaining readers are planned in `docs/plan-hosts-wave-2.md`... how much is ) + +- **Jobindex breaks your robots.txt rule.** Our search URLs use `sort=`, `page=` and `geoareaid=`, and Jobindex's robots.txt disallows all three. Search providers never read robots.txt, although you decided robots.txt is honoured. + +- Linkedin + + + +### fixes + +- Make sure all top have text if needed and the pipeline works including embedding. Also make sure they end in the top +- **Ranking by seniority.** You asked for a way to "sort out" seniority. There is a filter now, but the score still ignores the level. Since the column is filled for 43% of roles, a prefer gate on director and exec would now do something; none is set on your agent. +- **The top search box is unchanged.** It still matches title and company as plain text, so `Skane` or `Orsted` typed there finds nothing. Only the Company and Place chips fold letters. +- **Eightfold returns 10 roles a page and `MAX_PAGES` is 100**, so Infineon (1,289 roles) and Microsoft (about 2,300) would never be read complete and are not attached. The endpoint takes a location filter; a target has nowhere to state one. + +### Companies + +Get the company info for all, just add an info what the company is doing, sector, and then the extra info if we have it, like seed stage and the likes if we have it.. + +- Make a good company list and place list .. symbolized .. with aliases **`company_key` keeps `ø`**, so "Orsted" and "Ørsted" are two companies. Folding it changes the primary key of `companies`. **Place suggestions take about 110 ms**, most of it splitting 5,600 place lines into towns on every request. Caching the towns per distinct line would cut it. + +Startupmap get in + +#### Getting the companies + +You mentioned a startup map site that lists European startups with their funding stage and location, and said it would be good to have a company table with size, funding, recent hiring and tags, filled from such sources, with a button to refresh it every six months. You expected that the site would need a browser scraper because it is built in JavaScript. + + + +**startupmap.one**, the site you linked, lists about 4,500 European startups: + +| Field | Example | +| ------------------------------ | ------------------------ | +| Funding stage | seed, series A | +| Last funding amount | 19,230,769 | +| Team size, minimum and maximum | 60 to 70 | +| City and country | Barcelona, Spain | +| Sector | climate tech, healthtech | +| Hiring status | actively hiring | +| Year founded | 2020 | + +It also lists the jobs of those startups with their application links. + +**The Hub**, a Nordic startup job site, lists about 9,000 startups with employees band, year founded, funding stage, industries and countries. + +#### What "enrichment" would build + +1. **New columns on a company:** size band, year founded, funding stage, last funding amount, industry tags, plus the source and the date each fact came from. +2. **A job that fills them** by matching your companies against those two sources by name and domain. +3. **A refresh button**, and a schedule of every few months, so a new funding round is picked up. +4. **Filters on the feed and the company list:** for example "seed to series B, under 200 people, climate tech". +5. **A company view** when you click a company in the reader. + +#### Limits you should know + +- **Coverage is startups only.** Maersk, Novo Nordisk or LEGO are in neither source. For large companies the facts would come from elsewhere, or from a model reading the company's site. +- **Matching by name is imperfect.** "Bolt" or "Lunar" match several companies. Matching on the website domain is safer, and your seed stores the domain for every company. +- **Both sources are also a source of new companies.** A startup you do not follow yet, in a sector you like and marked as hiring, is a candidate to add. That is a second use of the same data. +- **Terms of use.** Both endpoints are public and unauthenticated, but they are not documented as an API for others. Reading them a few times a year for personal use is low impact. I would not poll them daily. + + + +## Frontend + +Can we make sure this make sense in a UX way. If we cant get that info then we need to make sure that the UI does not break + +- **"All jobs in Denmark."** The Denmark chip finds 1,193 roles. Another 2,152 open roles have no country, and their place line says only things like "Hybrid" (303), "Remote" (178) or "2 Locations". Some of those may be Danish and no filter can find them until extract reads the posting. +- **Remote, hybrid or on-site is known for 12% of Danish roles** (135 of 1,132). The Remote filter has the same blind spot you worried about for seniority, and it has no "and roles that do not say" box. +- **Setting Skåne on your agent.** I told you to enter `region:skane` under Search agents. The backend accepts it, and a plain "Skåne" typed as a town also matches the region, but I did not test that page's editor with it. + + + +## Fixes. + +- **Fit does not read seniority.** With three prompts on the live data of 2026-09-29 an individual contributor's posting ranks beside a head's on the same subject. Switch extract on so `seniority` fills, then add a prefer gate on it, and vote. +- **Votes do not train fit.** A vote only moves the rank at read time. Feeding votes into the prompt vectors is worth doing once there are a few hundred. +- **The boilerplate moved on.** Since roledesk's last refresh (`77052a8`, the schedules fix) it has `4573f6d` (the kit: the layout patterns every app built for itself, as classes) and `1679e88` (deploy from a WSL checkout, the sandbox setup). Refresh `deploy/` and `pipeline/` with `git archive`, and see whether `app/static/app.css` should use the kit's classes instead of its own copies of the same rules. +- **The live database has two agents**, "Berlin backend" and "Remote EU stretch", so Ranking edits the first and the Roles order menu lists both by name. Delete one with the CLI for a plain Best match. +- **Summarising is off** until the switch under Runs (default `summarise.enabled: false`) is used, and needs a model backend, as extract does. The Summarise button on a read page works either way once a backend is configured. +- **Nine Teamtailor seed companies have no board**, because their `careers_url` is the marketing site: Templafy, Queue-it, Telenor, Polestar, IQM, TeamViewer, Tractive, Starship, Factorial. +- Clean up the plans +- **The hosts added on 2026-09-29 were read from this machine only.** Every Workday tenant is behind Cloudflare; read one Workday and one SuccessFactors board from the deployed server before relying on them. +- **An unknown SmartRecruiters or Gem id answers an empty listing, not 404**, so a mistyped slug reads as a complete board with no roles and would close a company's roles. `set_board` reads the board first, which catches a typo only when the person notices "0 roles". +- **`Crawl-delay` is not honoured.** Booking.com's robots.txt asks for 5 seconds; the shared reader waits `PAGE_DELAY`, 0.5. A read of its board is two requests. +- **A SuccessFactors site on a company's own domain is not discovered from its own pages**, which link only the back end and the image CDN. It needs the rule Teamtailor has in `discover.py`: a page with the markers means the page's host is the board, checked by reading `/search/`. Until then such a board is set by hand. +- **HR Manager's portal disallows reading in robots.txt, its feed host has no robots.txt.** The module checks the feed's host, so a board would be read. Seaborg, the one company on it, is not attached until the owner decides whether the feed counts as allowed. +- **Six new hosts' dates are not trusted** (Eightfold, Jibe, Gem, Deel, HiBob, HR Manager; Pinpoint states none), so their backlog counts as posted when first seen. Check each date on a live board and add the host to `TRUSTED_POSTED_HOSTS`. +- **SmartRecruiters and Workable roles have no text.** Their lists carry none and their modules have no `detail`, so 7,391 open roles (Bosch, Wise, Ubisoft, Delivery Hero) are ranked on the title alone. Both hosts have a per-posting endpoint; a `detail` in each module fills the text 200 roles per board per pull. +- **A posting fetched by hand feeds the reader only.** "Fetch the posting" stores `article_text`, which `summarise` reads and `embed` and `extract` do not. +- **A list's teaser is not stored.** Phenom's and Oracle's lists carry a short description, which is dropped so that it cannot replace the full text; until the role's page is read the role has no text at all. +- **A starred company's roles crowd the top of Best first.** With the companies weight at 0.2 a favourite's roles gain about 0.13; after starring Databricks it held 33 of the top 60 rows. Lower the weight on the agent's card, or star fewer companies. +- +- + +## New stuff + +**Are the filters useful at this coverage?** Only with "no level" left on: + +- In agent 2's top 200, "manager and up" keeps 37 known and 70 unknown. Turn "no level" off and only 37 remain. +- Pay is filled on 12% of roles. +- 14 of the top 200 are interns or working students, because the agent has no rule on level. + +Align on location/ multiple locations + +Align on seniority + +Extra fields + +Make companies and locations a DB + +robot.txt + + + + + +**Bugs that bite now** + +- **Ashby marks hybrid roles as remote.** 2,537 roles pass every remote gate, and about 23 of OpenAI's 550 are really remote. +- **Lever keeps only the intro paragraph**, about 80% of the text lost. +- **A role with a place but no country is gated out.** For example, Zeiss in "Oberkochen; Jena" fails agent 2. This affects 1,262 roles. +- **Extract would mark a title-only role done for good.** It needs fixing before extract is switched on. +- **The live server is running without the duplicates change.** It started before that merge and hasn't applied the database migration yet. It needs a restart. + +clean up plans + + + +**1. Company facts and a company page (what you described).** + +- **Where it stands:** the columns were added in migration 20: sector, stage, size band, founded, last funding, and where and when each fact came from. All of them are empty on all 399 companies. +- **Startups:** read startupmap.one and The Hub. Both have open JSON endpoints and need no browser. Match companies by website domain, not by name, because names like "Bolt" or "Lunar" match several companies. +- **Big companies (Maersk, Novo, LEGO):** they are in neither source. My suggestion is Wikidata. It is free and open, and it has sector, headcount, year founded, headquarters and website. For the one-line "what they do", a model can read the company's own About page. +- **Company page:** click a company in the reader to open it. It shows the facts, open roles, board and reach, plus a Refresh button. A refresh on its own every few months is enough. +- **Filters:** stage, size and sector on the feed and on the company list. +- **Finding new companies:** the same data can suggest companies you don't follow yet, e.g. startups that are hiring in a sector you like. "Look at byfounders" from your list fits here: a VC's portfolio as another source of companies. + +**2. LinkedIn.** `decisions.md` already settles the method: alert mails over IMAP, no scraping. The pieces are: + +- read the mailbox +- parse each alert +- find the company with discover +- give it a board, or put it on the Watchlist through `add_by_hand` +- show the role in the feed, linking back to LinkedIn + +I need a mailbox and an app password from you. + +**3. Ranking quality.** This decides whether the top of the feed is actually good: + +- Fit still ignores seniority: 14 of agent 2's top 200 are interns or working students. A prefer gate on "at least rung X" would read the level the way the rail filter already does. +- A favourite company's roles crowd the top: after starring Databricks it held 33 of the top 60. +- Votes don't train fit yet. That's worth doing once there are a few hundred. +- Agent 2 accepts any role in DE and SE. Is that what you want, or only DK, remote and Skåne? + +**4. Coverage.** + +- 35 companies are readable but have no reader yet (Ørsted, Zalando, Amazon, Bolt and others). The plan is docs/plan-hosts-wave-2.md. +- Infineon and Microsoft need a location filter in the target, because of Eightfold's page cap. +- Nine Teamtailor companies still have no board, because their careers link points at the marketing site. + +**5. Bugs from your list to recheck.** The readers rework may already have fixed some, so each needs checking against current main first: + +- Ashby marks hybrid roles as remote. +- Lever keeps only the intro paragraph. +- Roles with a place but no country fail the gates (Zeiss "Oberkochen; Jena"). +- The top search box doesn't match `Orsted` or `Skane` without the special letters. + +**6. Housekeeping.** + +- Refresh `deploy/` and `pipeline/` from the boilerplate: it has moved on by two commits (the kit's classes, deploy from WSL). +- Delete the plans that are done or superseded. +- Decide on Jobindex's robots.txt, as you already did for Adzuna. +- Closed roles are never deleted, and their vectors grow by about 1 GB a year. +- Place suggestions take about 110 ms. +