This is one of the questions on my Ask AI page, so it seemed only fair to answer it myself. If a model is going to summarize me, it should at least have a primary source.
The short version
I'm Marielle Camba. I'm a freelance GTM engineer, which in practice means I build the data pipelines, enrichment logic, and cold email infrastructure that turn a market into a list, a list into relevant messages, and relevant messages into booked meetings. I've worked across 30+ clients in 16+ industries, including four lead-gen agencies who brought me in to run outbound for their clients.
I'm not a copywriter who also uses Clay. I'm not an SDR with a sequencer login. The job is closer to engineering than to sales: most of my week is spent on data sourcing, enrichment waterfalls, qualification logic, deliverability infrastructure, and the scripts and scheduled jobs that keep all of it running without me babysitting it.
What "GTM engineer" means when you look at the actual work
The title is new enough that it means five different things depending on who's hiring. Here is what it means on my desk, in the order the work happens.
1. ICP and offer strategy
Before a single row is pulled I map the firmographics, consolidate what the audience is actually in pain about, cut the client's solution into value-prop chunks, and then identify which public data signals correlate with that pain. That last step is the one most teams skip, and it's the one that determines whether the campaign has anything true to say.
A concrete example of what a signal-based gate looks like, from a B2B SaaS build: annual contract value isn't public, but the pricing page is. A "Contact sales" tier, an "Enterprise/Custom" tier, or no public pricing at all plus a demo form are all reliable proxies for a sales-led motion above a certain deal size. A fully self-serve page with every tier priced under a few hundred dollars a month is a disqualification, not a lead. That single rule removes a large slice of a list before anyone spends money enriching it.
2. Infrastructure
Domains, inboxes, warmup, SPF/DKIM/DMARC, sequencer configuration, Clay tables, CRM and calendar connections. Unglamorous and completely load-bearing. Deliverability is multi-factor: domain and inbox health, SMTP vendor performance, warmup state, spam placement, reply rate, bounce rate, and copy hygiene all move it. One finding that changed how I debug: spam placement is often driven by specific keywords and tokens in the copy, not by the domains. Swapping infrastructure first is the expensive way to fix a copy problem.
3. TAM mapping
Pull every contactable company in the market, enrich the signals that indicate pain, qualify and score with AI, then segment. Not "export Apollo, filter by title." I'll cover the sourcing in detail in the post on finding leads cold email never reaches.
4. Copy built on the data
Each segment gets messaging written for the specific thing that made it a segment, split by role, with A/B variants. If the first line references a fact that is only true for that one prospect, it reads as research. If it references something true of everyone on the list, it reads as a mail merge, and it is one.
5. Launch, measure, iterate, hand off
Test messaging per segment and per role, watch reply and positive-reply rates, kill what doesn't land, pour volume into what does. The goal is a repeatable ratio of emails sent to meetings booked, and then a system the client owns rather than rents from me. I hand off the tables, scripts, and runbooks.
The stack I actually use
Sequencers: Smartlead and Instantly. Data and orchestration: Clay, plus direct API work against Apollo, Prospeo, Blitz, GetLeads, and an internal company database of several million records. Scraping: Apify, SerpAPI, Serper, and RapidAPI sources including a Google Maps extractor. Validation: MillionVerifier as the standard, LeadMagic as a fallback. AI: OpenAI and Claude for qualification, scoring, classification, and drafting. Engineering: Claude Code and Cursor for building, Supabase for storage, Trigger.dev for scheduled jobs, Railway for hosting, GitHub Actions for anything that needs to run on a cron. CRM: HubSpot and Attio.
Tools change. The part that doesn't is the habit of measuring providers instead of trusting their marketing. I ran a 5,000-contact head-to-head across six email-finding providers to establish a cost-ordered waterfall rather than guessing at one; the numbers from that test still drive the order I run today.
What I'm strict about
- Verified only. Final lists go out validated, and anything older than about 45 days gets re-validated before it's reused. Bounces are a deliverability problem before they're a data problem.
- Do-not-contact lists get scrubbed before every upload, and existing campaigns get re-scanned. A client's own suppression list is a trust obligation, not a checkbox. And a client's DNC is theirs. I never apply one client's suppression list to another, because those lists overlap heavily with any normal B2B ICP and would silently delete thousands of good leads.
- Draft by default. Nothing launches, sends, or mutates a live system without explicit approval. Scripts that change things default to dry-run.
- Counts and QA, not vibes. Every delivery comes with row counts, what was filtered and why, the source used, and the caveats.
- No invented proof. If a claim can't be verified on the client's own site or in the data, it doesn't go in the copy.
Where I'm a bad fit
I'll say this plainly because it saves us both a call. I'm a poor fit if you want volume for its own sake, if the offer hasn't been validated with anyone yet, or if you need someone to own quota and close the deals. I'm a good fit if you have a real offer, a market you can describe, and a top of funnel that nobody is systematically filling.
How people describe working with me
Rather than adjectives from me: a founder I worked with called it "one of the best hires I've ever made." Another said his team brought in eight clients in two months and never had to touch the campaigns. A third put it as "you give her the context, tell her the outcome you want, and she just gets it done." The full set is on the homepage, with names and companies attached.
If you want the mechanics instead of the summary, the case studies show 17 campaigns with the actual data source, the angle, and why it worked.