sourcing tools keep giving me the same obvious people and missing the passive ones
I source for a pretty niche area, ml infra roles mostly. My problem with the tools ive tried, hireez and seekout, is they keep surfacing the same visible profiles, the people who already have open to work up or who show up in every recruiters search. Those…
I source for a pretty niche area, ml infra roles mostly. My problem with the tools ive tried, hireez and seekout, is they keep surfacing the same visible profiles, the people who already have open to work up or who show up in every recruiters search. Those people are already getting 20 messages a week. The ones i actually want, the engineer who hasnt touched their profile in two years but is doing the exact work, dont come up. Part of it is these tools lean on title and keyword matching, and the best people in a niche dont describe themselves the way the search expects. I want to search on what someone has actually done, past companies, the specific skills, github activity, rather than whether their headline happens to match. For people sourcing in a technical niche, what are you using that surfaces the passive candidates and not just the obvious ones. The title matching approach isnt cutting it for me
Collected discussion
You can do this as a Claude workflow with a couple of MCPs instead of leaning on one sourcing tool. This helps it be very custom to you, and Claude learns your judgment over time, so in my experience it makes it a lot more intuitive and recruits the way you do. The setup that i use is: Crustdata MCP - people search filtered on past companies, skills and tenure, so you pull people by what theyve built not their headline, gives you the full work history too GitHub MCP - for eng specifically, sourcing off real contribution and the repos surfaces people who never touch their profile FullEnrich - for contact info, they have a waterfall so its actually pretty accurate and really nice What i do is i'll just give Claude my brief, either my job description or a detailed brief. It'll search on multiple filters, maybe past company plus the skills required and the job experience, and check GitHub. It'll rank the list for me and personalize the outreach. I'll look at the outreach, review it, and send it out. For your use case with ML infra i think this kind of a setup probably makes a lot more sense, because normal recruiting tools cant just think like you. You need a lot of industry experience and that gut to make the right candidate choice.
Our sub is intended for meaningful discussion of recruiting best practices, not for self-promotion, affiliate links, or product research
Keep iterating your boolean string
Have you tried reaching out through niche forums or communities where these engineers might be more active?
I love conference speaker lists (and attendee lists if your company attended or plans to)
Wrangle mcp is pretty solid for niche engineering searches, just have claude/oai orchestrate
I'm currently having a pilot with a startup called Beaverhand out in Portland, they get applicants via google job search and run the assessments to verify their skills, I close 3 roles so far with them and I like it. We had to change our process in my firm, résumés and fancy LinkedIn profiles are kind of dead.
I use Sales Navigator for sourcing.
That usually means the search strategy is too keyword-dependent. I’d build searches around adjacent titles, competitor org charts, conference/speaker lists, certification groups, and project signals instead of only current title plus location.
You need to branch out on your Boolean string. I rarely title search
You're describing the core limitation of most sourcing tools. They're great at finding people who are easy to find and terrible at finding the ones you actually want. The engineer who hasn't updated their profile in two years but is doing exactly the work? Invisible to anything built on title and keyword matching. If someone doesn't maintain a polished profile on one network, they don't exist in the results. For niche technical roles, candidates leave footprints across multiple places. You need something searching across those surfaces rather than re-ranking the same index everyone else is querying. A few tools take that wider approach and pair it with human review before candidates hit your inbox. Worth looking into if title matching is consistently missing your target.
This comment was deleted.