Applied AI GTM company
Forward-deployed engineering and agents inside your Clay, CRM and sending domains, for AI startups from Seed to Series B. We aim at one number you pick and stay until it moves.
147 tier A follow-up tasks queued, 15 a weekday. X = not a fit.
Not a mock-up. Xeme's own engine, numbers from 14 October 2026.
Our own engine
The same kind of system we build for clients, running on Xeme's pipeline: HubSpot in, Clay research and scoring, verified email, a copy check, and everything written back with a reason.
Built on the platforms you already own
HubSpot
n8n
Claude
Supabase
NotionWhat we build
Each ends with a named process running in your systems, a person on the decision, and a number measured before and after.
Built on the tools you already pay for, one change at a time, tested on real records, then handed over with a runbook and an owner.
UNKAPPED: inbound qualification from 4 hours to under 3 minutes, 698 unit and 143 end-to-end tests.
A narrow job: read, decide, draft, write back. A score on every output; below the bar, a person. A test set from your past cases.
Relay for DefenderMate: Claude drafts every reply and nothing sends until a person decides.
Signals scored with a reason, accounts enriched once with a credit cap, routing with rules, every touch logged in your CRM.
RemoteState: qualification from 48 hours to 9.6, $180K ARR growth in six months.
One department process at a time: baselines first, then what each agent reads and writes, one change a week, a weekly review.
Most production work so far is in revenue teams. The live system below is the clearest example.
Live in production
Built in-house by Rasul at UNKAPPED. Every recorded outside call becomes one record tied to the right client, with a summary a person reads first, and nothing reaches a client or the CRM without a gate.
Calls came in from 7 recorders, and the same call often arrived more than once. 179 transcripts turned out to be 109 calls.
Every recording is saved once, merged into one call, tagged by type, matched to the client in HubSpot (read only) and summarised. Calls the system is unsure about go to a person.
All 29 workflows live, 9 of them switched on today. In the last 7 days 638 of 653 runs finished clean. 47 summaries across 39 calls, and 22 questions put to the call library, 16 answered with quotes. 1,620 tests pass.
Task drafts into one team board (at most 15 new and 30 changed tasks a day, confidence floor 0.6), review buttons, a pre-call brief, the pilot digest, calendar gaps, Gemini notes and the daily health check. Nothing is written to HubSpot.
Read from the live n8n and the database on 27 Oct 2026.
How a Sprint runs
A build schedule, not a discovery phase. About two hours of your time in week one, plus access to your CRM and domains.
The outcome and the guardrails. TAM, ICP, personas and signals agreed before anything is wired.
Domains, mailboxes, warm-up, SPF, DKIM, DMARC, CRM fields. Signal feeds go live.
Engines run against past replies, lost deals and closed calls before a live mailbox. You correct them.
First qualified meetings. Inbound scored and routed. Copy follows the reply data.
Attribution in your CRM, runbooks, handover. The engine keeps running in your accounts.
Research, draft, check with a second model, route, follow up, and log every step to the CRM.
Set the rules, correct the first runs, decide anything unusual, and take the meetings.
Inside one engine
The personalization engine Rasul built in-house at Omnibound AI: 95%+ deliverability and 1.5%+ positive replies across 150+ sending domains.
AI search
Buyers ask an assistant for a shortlist before they visit your site. We measure how often ChatGPT, Claude, Perplexity and Google's AI Mode name you, fix the gaps, and use each gap as a reason to write.
Real buying questions from search data, run many times on each engine, because answers change from run to run.
Who gets named, which sites get cited, where you are missing. Each engine cites different sources.
Direct-answer pages with numbers and sources, comparison pages, and mentions on the review sites, lists, Reddit threads and videos assistants cite.
The panel re-runs monthly and gets its own line next to pipeline. Nobody can promise placement; we report the trend.
At Omnibound AI, Rasul built 500+ personalized ABM pages on this signal, one per account, each visit logged.
Test it on us. Ask an assistant what Xeme does:
Pricing
Everything that ships runs in your accounts and stays yours.
A paid audit of your GTM stack and signals, delivered as a node-by-node build plan.
The whole engine: plan, signals, personalization, sending setup, CRM tracking and routing.
Running the engine after a Sprint: new signals, campaign changes, ABM pages, monthly attribution review.
Add-on modules, priced per build: AI search visibility, an agent for one process outside sales, founder content for LinkedIn and X, speed-to-lead routing.
Proof
Systems Rasul built inside each company, before or alongside Xeme. Not Xeme client results. Details on request.
The stack
Nothing gets migrated, and models stay swappable. The dashed layer is what agents connect to when a process needs it.
Claude and Claude Code
Anthropic API
Gemini
n8n
PythonSupabase
PostgreSQL
Modal
Vercel
GitHubClay
HubSpot
Fathom
Semrush
Perplexity
Search Console
GA4
Zendesk
Intercom
JiraNotion
Airtable
SAP
Snowflake
Databricks
Google CloudAsk which of the dashed layer we have shipped on, and we will tell you.
Who runs it
Rasul Shaikh. AI GTM engineer since 2020, founding forward-deployed engineer at UNKAPPED, official Clay Creator, 3x Clay Certified. Pune, works remote.
You bought the AI. You bought the GTM tools. The number still hasn’t moved. That’s the gap I work in.
His day job is at UNKAPPED; Xeme is the practice he runs alongside it, so it takes on a few builds at a time.
Questions
Venture-backed AI and ML companies from Seed to Series B with a sales-led or hybrid motion: founders still doing founder-led sales, first GTM hires, or revenue leaders whose SDR math stopped working. Not a fit: pure PLG with no sales motion, or teams that want rented SDRs.
An engineer who works inside your team and your systems, maps how the work actually moves, builds on the tools you already pay for, and stays until it runs in production. Then hands it over with a runbook and a named owner on your side.
AI tied to one named business process, running on your own data and systems, with an owner, a review gate and a number it is judged on. Buying seats for a chat tool is not applied AI.
Building pipeline as a system rather than a campaign: signals captured and scored with a reason, accounts enriched once, routing with rules, every touch logged in the CRM, and closed deals feeding back into targeting.
We can measure it and move it, but nobody can promise it. Answers change from run to run and each engine cites different sources, so we track a panel of real buying prompts every month and report the trend next to pipeline.
You decide. Outbound email can go out after a second model checks it, once you have approved the rules and the first runs. Replies to prospects, anything that touches money, and anything below the confidence bar wait for a person.
About two hours of founder or GTM-lead time in the first week, access to your CRM and domains, and sign-off on the signal plan. After that you review the first runs and take the meetings.
Yes. It is built in your Clay, your CRM, your n8n and your domains. When the work ends, the engine stays and keeps running, with runbooks and handover docs.
Usually that is the most useful case. The Teardown maps what to keep, and the Sprint rebuilds the engine around it. Existing tools shorten the build.
Next step
Bring the number you want to move. If Xeme is not the right fit, we will say so and point you to someone who is.
Replies within one business day. ops@xeme.co
ChatGPT
Claude
Google AI Mode
Perplexity
Grok