ALEX KEDRA
Sourcing Playbook
alexkedra@tuta.io
Forward Deployed Technical Recruiter

How I find people
most recruiters never see.

This document explains how I actually work — not the version that fits a job description, but the operational system I've built over four years of hiring in some of the scarcest technical talent markets in the world: cryptography, ZK protocol engineering, DeFi infrastructure, and now AI.

19 Specialist roles · zkSync
~90% Outbound response rate
45–60d Avg time-to-hire · scarce markets
50+ Engineers hired globally
01 · Philosophy

Recruiting is a
signal problem.

Most recruiting processes are optimized for volume. More outreach, more applications, more pipeline. That logic works when the talent pool is large and the role is well-understood. It fails completely when you're hiring a ZK protocol engineer, a cryptographer, or a backend engineer who has built trading infrastructure on open protocols.

In those markets, the best people aren't on LinkedIn. They don't apply to jobs. They're skeptical of recruiters. And they can tell within two sentences whether you understand what they actually do.

The job isn't to run a better process. It's to develop better signal — about the role, the candidate, and the ecosystem — before a single message gets sent.

Embed before you source. I don't start sourcing until I understand why the last hire failed, what's genuinely non-negotiable, and what the team's actual working style is. This changes everything downstream.
Map the ecosystem, not the job board. Every technical talent pool has a geography — forums, GitHub orgs, Discord servers, conference circuits, event attendee lists. Find the geography first.
Design screening for the role, not the template. Generic screening questions produce generic candidates. The right question for a protocol engineer is different from the right question for a trading infrastructure engineer.
Outreach is a test of your understanding. If you can't write one specific sentence about why this person's work is relevant to this role, you haven't done enough research to send the message.
02 · Process

How a search
actually runs.

This is the sequence I run for every new search, regardless of role type. The specifics change. The structure doesn't.

01

Embedded kickoff — not a brief

I don't start with a job description. I start by understanding the team's failure modes: who didn't work out, why it took so long to figure that out, what was compromised on in previous hires, and what the non-negotiables actually are versus what sounds good on paper.

02

Ecosystem mapping before sourcing

Before opening any sourcing tool, I map where the target population actually lives — GitHub organizations, research forums, Discord communities, conference speaker lists, event attendee networks, open source contribution graphs. The sourcing strategy follows from this map.

03

Signal definition — what does good actually look like?

I work with the hiring team to define what genuine competence looks like in this specific role. Not generic seniority signals, but role-specific indicators: contribution patterns, problem types, depth of open source work. This defines the sourcing filter before I look at a single profile.

04

Targeted outreach — specific, not scaled

Every outreach message references something concrete about the person's work. This is slower than mass outreach and produces dramatically higher response rates — consistently 85–90% across searches in skeptical-of-recruiters talent pools.

05

Recruiter screen — designed for the role

I build custom screening questions for each search that filter on genuine depth before engineering time is spent. The goal is to make the process respectful of the candidate's expertise while giving the team high-confidence signal before the technical loop begins.

06

Pipeline management and stakeholder rhythm

Structured weekly syncs, pipeline health tracked against agreed metrics, and problems flagged early. If a search is going to take longer than expected, I say so at week three, not week ten.

03 · Case Study

Matter Labs (zkSync)
19 specialist roles.

zkSync is one of the leading Ethereum Layer 2 scaling solutions, built on zero-knowledge proofs. Hiring for this team meant finding people who understood ZK cryptography, distributed systems performance, and protocol design at a level where the global candidate pool is measured in hundreds, not thousands.

The Problem
Standard recruiting channels returned near-zero results.
Protocol engineers with genuine ZK proof system knowledge don't apply to job postings. Most aren't active on LinkedIn. Many are academics or open source contributors who have never engaged with a recruiter. The talent pool is global, niche, and deeply skeptical of anyone who doesn't demonstrate immediate technical context.
Ecosystem Mapping

I built sourcing infrastructure around where this community actually lives: Ethereum Magicians forum, ZK-focused GitHub organizations, academic paper authorship on ZK proof systems, conference speaker lists (ZKSummit, ZKProof), and contribution graphs on open source cryptography libraries.

Beyond sourcing channels, the screening process needed to change. I designed recruiter-led screening questions calibrated to algorithmic depth — questions I could ask in a first call that gave the engineering team confidence before they invested interview time:

"What is the time complexity of binary search, and can you order these from slowest to fastest: O(log n), O(√n), O(n)?"
Why this question: Simple enough for a recruiter screen, specific enough to reveal genuine CS fundamentals. Candidates who hesitate on the ordering reveal they lack the depth required for protocol-level work.

✓ Strong signal: Instant recall, correct ordering (log n → √n → n), often extends into follow-up reasoning unprompted
✗ Weak signal: Correct answer but slow, uncertain on ordering, can't explain the underlying logic
19Specialist roles filled
~90%Outbound response rate
45–60dAvg time-to-hire
50–100+Qualified candidates per critical role
2 moTo first protocol hire after new screen
04 · Case Study

0x / Matcha —
embedded from day one.

0x builds open infrastructure for decentralised exchange. Matcha is their consumer-facing DEX aggregator. The roles were technically specific in ways a standard job description couldn't capture: backend engineers who understood trading protocol internals, search infrastructure outside the Google ecosystem, and mobile engineers who had shipped real Web3 trading products.

Week One
I didn't post a job. I joined the engineering standup.
Week one wasn't sourcing. It was diagnosis. I embedded directly with the engineering team — daily standups, product reviews, direct conversations about what they were building and where the real gaps were. Before any outreach started: why had previous searches stalled? What had been compromised on? What did the team actually need versus what was in the JD?
What the JD said
What the team actually needed
Backend Engineer
TypeScript, Node.js, backend systems experience
Real signal
Deep understanding of trading protocol mechanics, order routing, latency constraints. OpenSearch or Elasticsearch experience — real search relevance engineering in financial data contexts.
Mobile Engineer
React Native, mobile app experience
Real signal
Had to have shipped a mobile trading product in Web3. Wallet integrations, transaction signing, real-time price feeds, DeFi UX patterns.
Protocol knowledge
Familiarity with DeFi protocols preferred
Real signal
Genuine understanding of how AMMs, order books, and DEX aggregation work. Engineers who had contributed to or built on well-known open protocols.
The Output

Active interview processes across 7 concurrent engineering roles within weeks of starting. ~85–90% outbound response rate. 40–60% conversion from outreach to interview. All pipelines built from zero.

05 · Sourcing

Where the talent
actually lives.

When standard channels don't work, the question is: where does this community actually congregate? The answer is different for every technical domain. Here's the ecosystem map I've built across the roles I hire for most.

Protocol / ZK / Crypto
  • Ethereum Magicians
  • ethresear.ch
  • ZKSummit speakers
  • ZKProof community
  • Academic ZK papers
  • privacy-scaling-explorations
  • L2 ecosystem Discords
Trading / DeFi Infra
  • Protocol contributors (Uniswap, 0x, Aave)
  • DEX tooling GitHub orgs
  • Web3 trading Discord servers
  • DeFi-focused hackathons
  • OpenSearch / ES community
  • ETHGlobal winners
  • On-chain contributor history
AI / ML Engineering
  • Hugging Face contributors
  • ArXiv authors (applied ML)
  • Inference / training infra GitHub
  • Papers With Code leaderboards
  • NeurIPS / ICML attendees
  • ML engineering Discords
  • AI Twitter / X clusters

The ecosystem map comes before the sourcing tool. Understanding which community to enter — and how to enter it credibly — is most of the work.

06 · AI Tooling

Nine tools.
Built to find signal.

The manual version of ecosystem sourcing works. The AI-assisted version works at a different scale and with more precision. I've built a suite of recruiting intelligence tools that layer AI on top of four years of technical ecosystem knowledge — eliminating low-signal work so the high-judgment work gets more time.

Candidate Analysis
01
Candidate Momentum Tracker
Analyzes career trajectory, not just current role. Surfaces whether someone is accelerating, plateauing, or quietly ready to move — before first contact. Outputs momentum score, trajectory analysis, readiness-to-move assessment, and outreach timing recommendation.
Claude LinkedIn GitHub
02
GitHub Intelligence — Batch
Paste up to 10 GitHub profiles at once. Produces individual assessments covering technical depth, domain fingerprint, contribution patterns, and outreach angle — plus a ranked outreach priority list across the batch. Replaces 2–3 hours of manual profile review per sourcing sprint.
Claude GitHub Instant Data Scraper
03
Outreach Personalization Engine
Takes candidate intelligence and role context, generates a personalized first message referencing specific signals from their work — a repo, a conference talk, a forum post, a contribution pattern. Outputs primary message, alt version, follow-up, and what to avoid. Specificity at scale.
Claude GitHub intel output Ashby
Team & Market Intelligence
04
Team Composition Analyst
Analyzes a company's current engineering team before the kickoff call. Produces seniority distribution, skill coverage gaps, background concentration risk, recommended next 3 hires with reasoning, red flags in current structure, and questions to ask in kickoff. Walk in knowing more than the founder expects.
Claude LinkedIn Company site
05
Compensation Intelligence Builder
Synthesizes market signals into a structured comp brief with reasoning — not just a number, but what the candidate will compare it against, how to frame equity at this stage, negotiation dynamics, and how to position the offer. Outputs a briefing document for founder and HM before the offer call.
Claude Levels.fyi signals Market data
—
Clay Enrichment Workflow
Takes a list of GitHub usernames or LinkedIn profiles from target ecosystems, runs automatic enrichment across both platforms, pulls recent activity signals, and outputs a scored, prioritized candidate list with context ready for outreach. No manual cross-referencing. Ecosystem mapping to ranked list in one workflow run.
Clay GitHub API LinkedIn
Ecosystem Sourcing
06
Luma Attendee Analysis
Technical events publish attendee and speaker lists via Luma. I extract these lists and pass them through Claude to produce a talent map: who attended, what they work on, priority tier for outreach, and a specific hook from their event presence. Turns a conference attendee list into a sourcing pipeline within an hour.
Claude Luma LinkedIn enrichment
07
Conference Speaker Intelligence
Speaker lists are the highest-signal sourcing data available — community-validated experts. I scrape speaker lists from target conferences and pass them through Claude with role context to produce assessments of relevance, specific outreach hooks from their talk content, and an approach strategy that respects their community standing.
Claude Web scraping Clay enrichment
08
Community Cluster Mapper
Maps the talent network in a specific technical community from seed accounts and topics. Identifies community geography, high-signal sourcing nodes, where non-obvious talent lives, and produces a 48-hour sourcing sprint plan. Surfaces people who are intellectually active but not yet on everyone's radar.
Claude X / Twitter GitHub orgs
Pipeline Ops
09
Rejection Pattern Analyzer
Treats rejected candidates as signal, not dead ends. Analyzes rejection data to find where searches stall, separate sourcing problems from evaluation problems, surface interviewer consistency issues, identify re-engagement candidates, and produce an action plan for the next search. Includes root cause diagnosis and hiring team questions.
Claude ATS data Ashby export
→
The Principle Behind All of It
The goal isn't to automate recruiting. It's to use AI tooling to eliminate low-signal work — generic outreach, manual profile triage, repetitive research — so the high-judgment work gets more time: understanding the team, defining real hiring signal, and evaluating people accurately. The tools require ecosystem knowledge to use well. That's the moat.
4 years ecosystem knowledge
Building a technical team
in AI or Web3?

I'm looking for founding or first-recruiter roles where the work is building the system, not just filling the pipeline. AI infrastructure, DeFi, and frontier technical teams.

Get in touch →