I need to hire a Staff Machine Learning Engineer for our AI agents team. They need deep Python, production LLM application experience, production ML systems, and ideally search, ranking, recommendation, retrieval, or agent-infrastructure experience. We prefer Seattle or the Bay Area, but remote is possible.

Complete this workflow in order:
1. Calibrate: define the role mission, must-haves, nice-to-haves, title variants, and target-company segments.
2. Search strategy: specify the strict starting criteria and which variables can be broadened later without lowering the technical bar.
3. Search: run the SeekOut search using the strict criteria first.
4. Verify: evaluate candidates using profile evidence only. Do not infer a skill from title, employer, school, or industry; mark unsupported areas "unknown."
5. Rank: return the 10 strongest unique candidates with current title/company/location, fit tier, 3-5 evidence-based match signals, key gaps/unknowns, and one suggested first-call question.

If the strict search does not produce 10 strong candidates, broaden one variable at a time and tell me exactly what changed.

Find rising-star AI infrastructure engineers with credible evidence across SeekOut public profiles, GitHub, academic work, or other connected professional sources.

Focus on evidence related to LLM systems, retrieval, evaluation, distributed inference, serving, agent infrastructure, or adjacent production AI systems.

Evidence rules:
- Favor direct technical evidence: relevant repositories/contributions, publications, patents, technical talks, shipped systems described in profiles, or clearly documented production work.
- Do not treat GitHub activity volume, employer prestige, school prestige, or job title alone as proof of fit.
- Deduplicate the same person across sources before ranking.
- Clearly distinguish confirmed evidence from reasonable inference and from unknowns.

Return the top 20 candidates in a table with: candidate, current role/company, strongest evidence source(s), relevant technical signals, evidence-strength rating, likely seniority, and key unknowns.

Then identify the 5 candidates whose evidence is strongest relative to how obvious they would be in a title-only search.

I like the candidate profile currently in context because of their startup experience, strong backend-systems depth, and recent LLM product work.

First, reverse-engineer the profile into 4-6 job-relevant fit dimensions. Separate:
- Essential characteristics that should drive the search.
- Helpful but non-essential characteristics.
- Incidental traits that should not be copied, such as exact employer, school, or identical title.

Then find 15 similar candidates using the essential characteristics. Diversify the slate across companies and backgrounds; avoid more than 2 people from the same employer unless the evidence is exceptional.

For each result, show the 3 strongest similarity signals, the most important difference from the reference candidate, and any evidence gap. Base similarity on documented skills and experience, not superficial biography.

Use this hiring-manager feedback to recalibrate the current search:
- Candidate 1 is too enterprise SaaS.
- Candidate 2 is strong because of hands-on infrastructure depth.
- Candidate 3 is too research-heavy.
- Candidate 4 is close but too junior.

Before running the revised search, translate that feedback into explicit search rules:
1. What criteria should become more important.
2. What criteria should become less important.
3. What should stay unchanged.
4. Any new exclusion or seniority signals, stated carefully enough to avoid filtering out good adjacent talent.

Then run the refined SeekOut search and return 10 better-fit candidates. For each, show why they better match the updated calibration, the supporting evidence, and any remaining gap.

End with a short "what changed" summary comparing the revised slate to the previous one.

Build a ranked shortlist of the top 5 candidates for the role currently in context.

Evaluate everyone against the same job-related criteria already established in this chat. Do not give credit for a capability unless there is supporting profile evidence; label missing information "unknown."

For each candidate include:
- Rank and fit tier.
- One-line hiring-manager summary.
- Why they fit.
- 3-5 specific pieces of evidence.
- Concerns, gaps, or unknowns.
- 2 suggested interview focus areas.
- What would have to be true for this person to become the #1 candidate.

End with a short comparison of the top 5 explaining the key tradeoff between them rather than repeating each profile summary.

For the top 5 candidates in the current slate, draft personalized outreach using only job-relevant details supported by their profiles.

For each candidate provide:
1. LinkedIn message: approximately 250-400 characters.
2. Email: subject line plus approximately 90-120 words.

Rules:
- Use one specific personalization hook from the candidate's documented background.
- Explain the relevance of the role in concrete terms.
- Do not invent projects, interests, motivations, or achievements.
- Avoid generic AI language and exaggerated flattery.
- Keep the call to action conversational and low-friction.
- Do not reference sensitive or protected personal information.

If a candidate lacks enough evidence for meaningful personalization, flag that and write a role-specific version without fabricating a hook.

Map the AI platform and applied ML talent at Anthropic, OpenAI, Google DeepMind, Meta, and Databricks.

Analyze each company at the aggregate level across:
- Common relevant titles and title families.
- Seniority distribution.
- Concentration of LLM infrastructure, retrieval, evaluation, inference, distributed systems, and applied-ML skills.
- Geographic distribution.
- Likely feeder companies based on career-history patterns visible in the data.
- Candidate segments that may be more practically reachable for a startup based on role adjacency, career stage, geography, and company-tenure patterns - not assumptions about an individual's willingness to leave.

Use a comparison table first, then summarize:
1. The 3 most attractive talent segments for a startup.
2. The companies most likely to produce direct matches versus adjacent matches.
3. Search angles that could uncover less-obvious talent.

Call out data limitations where the available profile data does not support a confident conclusion.

Compare Seattle, San Francisco/Bay Area, New York, Toronto, and London for Staff-level ML infrastructure engineers with strong Python, distributed systems, and production LLM experience.

For each market, show:
- Estimated relevant talent-pool size, if available.
- Concentration of the required skills.
- Top employer segments.
- Common titles.
- Relative competition for this talent.
- Likely sourcing advantages and constraints.

Then score each market from 1-10 on:
1. Qualified talent depth - 35%.
2. Required-skill concentration - 25%.
3. Competitive intensity / recruiting difficulty - 20%.
4. Title and background diversity - 10%.
5. Practical access for this role - 10%.

Show the scoring logic, identify any data that is directional rather than measured, and recommend the best primary market plus the best secondary market.

Find academic and industry experts in AI evaluation, LLM benchmarking, and agent reliability.

Prioritize direct evidence such as relevant publications, benchmark or evaluation-framework contributions, conference participation, technical leadership, open-source work, or documented industry systems.

Rank experts on:
- Relevance to AI evaluation / benchmarking / agent reliability.
- Depth and recency of work.
- Evidence of translating research into practical systems.
- Industry-transition signals such as prior industry work, applied collaborations, startup experience, or product-facing research - without assuming personal interest in changing jobs.

Return a ranked expert map with: current affiliation, focus area, strongest evidence, recent relevant work, industry adjacency, reachability indicators, and key unknowns.

Do not equate publication volume or institution prestige with fit unless the work itself is relevant.

Analyze internal bench depth for three critical roles: Principal ML Engineer, Staff Backend Engineer, and Product Manager for AI Recruiting.

For each role:
- Define the 4-6 most important job-related capabilities.
- Identify internal employees with meaningful overlap.
- Group them by readiness: ready now, plausible successor with development, or partial overlap.
- Show where capability coverage is concentrated in too few people.
- Identify the largest bench gaps and the likely development path to close them.

Then rank the three roles by continuity risk if an incumbent or key contributor were to leave, based on bench depth and skill concentration - not on predictions about any individual's likelihood of leaving.

Support each readiness assessment with available evidence and label unsupported areas as unknown.

Give me a pipeline-health overview for our active engineering requisitions.

Break down the pipeline by:
- Requisition.
- Stage.
- Recruiter / owner.
- Candidate count.
- Stage age and time in process.
- Rejection or withdrawal reason where available.

Identify bottlenecks and stale stages using our configured SLAs or historical stage-duration norms if available. If neither exists, show the raw age distribution and label any outlier rule you use rather than inventing a company standard.

Highlight:
1. Reqs with the largest stage-conversion problems.
2. Candidates or stages needing action now.
3. Repeated rejection/withdrawal patterns.
4. Workload imbalances across recruiters if supported by the data.

End with a prioritized "intervene this week" list: action, owner, reason, and expected impact.

Analyze the current talent pool for sourcing concentration and coverage gaps, then recommend ways to broaden it without lowering the job-related bar.

Assess whether the search is overconcentrated in particular:
- Employers or employer types.
- Titles.
- Geographies.
- Schools or traditional credential paths.
- Career paths / industries.
- Professional communities or talent ecosystems.

If the connected system provides approved aggregate representation metrics, you may summarize those at the aggregate level. Do not infer or classify an individual candidate's race, ethnicity, gender identity, religion, disability, age, sexual orientation, or other protected characteristic from names, photos, schools, affiliations, or proxies.

Then recommend alternate titles, adjacent companies, geographies, schools/programs, professional communities, and non-obvious career backgrounds that could expand sourcing coverage while preserving the same qualification bar.

Prioritize the 5 expansion moves most likely to add qualified talent and explain the tradeoff for each.

For a Senior Product Manager, AI Recruiting role based in Bellevue, give me a current compensation and talent-market reality check. Compare Bellevue/Seattle, San Francisco/Bay Area, New York, and US remote.

Use the most current compensation or market data available through the connected tools. State the date or recency of the data and clearly label estimates or low-confidence ranges.

For each market compare:
- Typical base-salary range if available.
- Total-cash and/or total-comp context if available.
- Relevant talent-pool depth.
- Competition for AI product talent.
- Location-specific recruiting constraints.

If I have already provided our compensation range in this chat, compare it directly with the market. If I have not, show the market ranges first and tell me what employer range would likely be competitive rather than inventing our range.

End with: market difficulty rating, where we are most likely to struggle, and 3 levers besides compensation that could improve our odds.

Find cardiologists with California license signals and experience in heart failure and/or electrophysiology who are affiliated with major hospital systems.

Start with the Bay Area and Los Angeles. Then show how the qualified pool changes when expanded statewide.

For each candidate include:
- Current title and hospital/system affiliation.
- Primary location.
- California license signal and source/status if available.
- Heart-failure or electrophysiology evidence.
- Academic/clinical leadership signals if relevant.
- Key unknowns requiring verification.

Treat license data as something to verify when current status is not explicitly confirmed. Do not infer a specialty solely from hospital affiliation or department name.

End with the size and quality difference between the metro-only pool and statewide pool, plus your recommendation on whether statewide expansion is necessary.

Find nurse practitioners in Texas with emergency medicine and/or urgent care experience.

Start with candidates who have:
- Texas license signals.
- Direct NP experience in emergency medicine, urgent care, or closely related high-acuity ambulatory care.
- Recent clinical practice evidence.

For each result show: current role/employer, location, license-state evidence, specialty evidence, likely current setting, and any verification gap.

If the strict pool is too small, broaden carefully and one step at a time. Possible expansion steps are adjacent acute-care settings, nearby geographies within Texas, or related NP specialties with demonstrably transferable experience. State exactly what changed at each step.

Do not present a license as current or active unless the available data confirms it.

Find Federal Account Executives in the DC metro area with experience selling into federal civilian and/or defense agencies, plus security-clearance evidence or clearance-related signals.

Prioritize candidates with:
- Documented federal quota-carrying sales experience.
- Named federal civilian or defense customer segments, when available.
- Experience at defense contractors, federal SaaS vendors, cloud providers, cybersecurity companies, or adjacent federal-technology firms.
- Security-clearance information when explicitly available.

For each candidate show: current role/company, DC-area location signal, federal-sales evidence, agency/segment evidence, clearance status if explicitly documented, clearance-related signals if not verified, and key unknowns.

Do not describe a candidate as having an active clearance unless the data explicitly supports that. Do not infer clearance solely from employer, project, or geography.

Search our internal talent for employees who could move into an AI Solutions Engineering role.

Evaluate only job-related evidence. Look for:
- Customer-facing technical experience.
- Python, ML, data, APIs, or adjacent technical exposure.
- Strong product judgment and troubleshooting ability.
- Recruiting-domain or HR-tech knowledge where relevant.
- Evidence of cross-functional communication and ownership.

Group candidates into:
1. Ready now - evidence supports near-term transition.
2. Ready with targeted training - strong adjacent fit with specific, addressable gaps.
3. Longer-term / exploratory - interesting overlap but multiple material gaps.

For each person, explain the evidence, the main gap, and the specific training or experience that would improve readiness.

Do not use or infer protected characteristics, medical information, leave history, or other non-job-related personal information in the assessment.

Search our ATS for past candidates who could fit a Senior Backend Engineer role today.

Prioritize candidates who:
- Reached onsite, final, or another late stage.
- Have a documented prior disposition related to timing, compensation, headcount, location, or another non-skill factor.
- Have gained relevant backend, distributed-systems, platform, or leadership experience since the prior process.

Rank the strongest rediscovery candidates. For each include:
- Prior stage and date.
- Documented disposition reason; if unavailable, say "unknown."
- What has changed in the candidate's experience since then.
- Current fit evidence.
- Why re-engaging now could make sense.
- Suggested re-engagement angle.

Do not infer why someone was rejected if the ATS does not explicitly contain the reason.

For the current top 20 candidates, prepare the contact-and-export workflow.

Target workspace: "Staff ML Infra Q3."

Sequence the work like this:
1. Confirm the 20 candidates in scope and remove duplicates.
2. Add them to the workspace if they are not already there.
3. Before retrieving contact details or using credits, summarize how many lookups would be attempted and ask for my explicit approval.
4. After contact retrieval is approved and completed, summarize what was found.
5. Before exporting or sending data to our ATS, show the candidate count, destination, and fields to be exported, then ask for my explicit approval.
6. Export only after I approve.

At the end, report completed actions, skipped/failed records, and any candidates requiring manual review.

We need to hire a Staff Machine Learning Engineer for an AI recruiting agents team. The person needs production LLM experience, strong Python, distributed systems depth, search/ranking/recommendation or retrieval experience, and strong product instincts. Seattle or the Bay Area is preferred; remote is acceptable for exceptional fits.

Run this as a phased recruiting workflow in one chat:

PHASE 1 - CALIBRATE Define the role mission, hard requirements, ranking signals, title variants, and target-company segments. Keep hard gates separate from preferences.

PHASE 2 - MARKET Compare Seattle, Bay Area, New York, and Toronto on relevant pool depth, skill concentration, competition, and practical recruiting difficulty. Recommend a primary and secondary market.

PHASE 3 - SEARCH Search the relevant SeekOut sources, including public, GitHub, and academic signals where useful. Start strict, deduplicate identities, and broaden only one variable at a time if needed.

PHASE 4 - EVALUATE Build a ranked shortlist of 10. For each candidate show evidence supporting the match, gaps/unknowns, fit tier, and first-call focus. Do not infer skills from employer/title alone; label missing evidence "unknown."

PHASE 5 - PRIORITIZE & OUTREACH Identify the 5 candidates I should contact first and explain why. Draft concise, evidence-based personalized outreach for those 5 without inventing personal details.

PHASE 6 - EXECUTE WITH APPROVAL GATES Create a workspace for the approved shortlist if the tool supports it. Before any action that uses credits, retrieves contact details, or exports/sends candidate data to an ATS, summarize the action and ask for my explicit approval. Do not perform those gated actions until I approve.

At each phase, keep the output decision-oriented and carry forward the criteria established earlier in the chat.