Proof of work, not pitch decks.

Shipped systems, shown the way they actually run: the interface, the pipeline behind it, and the stack it sits on.

one shipped · six scopedone featured · the rest expandableJump to the index ↓
01 / Featured case

Lead Intelligence Agent

shipped · v1.4
internal + client deploys
lead-intelligence-agentrun #2,417 · 312 accounts scored · 41s
Account / signalFit scoreStage
Harborline Logistics
hiring 3 ops eng · series B · 140 hc
SENT
Cedarpoint Health
4 CRM tools detected · manual intake
QUEUED
Northgate Supply Co.
new RevOps lead · 60d in seat
DRAFT
Vantage Print Group
no buying signal · below threshold
SKIP
Generated draft · Harborline
subj: your 3 ops eng reqs

Saw you're hiring three ops engineers. Half that scope is routing and status chasing. We've automated it before in 9 days. Worth 20 minutes?

tone: peer42 words
Sequence
  1. 01 openersent d0
  2. 02 proof linkd3
  3. 03 teardownd7
  4. 04 close loopd12
reply rate11.4%

What it does

Pulls accounts from Apollo, enriches them against firmographic and hiring signals, scores fit with a Claude-driven rubric tuned per ICP, writes a first-touch email in the founder's own voice, and pushes the sequence to Lemlist with replies synced back to HubSpot. One operator reviews; nothing sends unreviewed.

Stack used

  • Claude API
  • Next.js
  • Node
  • Postgres
  • Apollo
  • Lemlist
  • HubSpot
  • Vercel
02 / System log

What gets built next.

Six systems already scoped, with architecture, sequence and stack worked out. Filter by the team that owns the bottleneck. Open a row for the detail.

What it does

Every inbound ticket is classified on arrival, routed by your own escalation rules, and answered with a draft written against your resolution history. Agents move from authoring to approving.

  1. 01Taxonomy and escalation rules derived from historical tickets
  2. 02Retrieval over past resolutions, with an eval suite per intent
  3. 03Draft attached to the ticket; agent approves, edits or escalates

Stack

  • Claude API
  • Next.js
  • Postgres
  • pgvector
  • Zendesk API

Before

Every ticket read cold, routed by hand, answered from scratch.

After

Classified and drafted on arrival. Humans hold the approve seat.

Ask about this one →

Filter shows 6 of 6 scoped systems · the shipped case above makes 7.

03 / FAQ

What people ask about the work.

How many systems has ANTA actually shipped?

One is in production: the Lead Intelligence Agent, built for ANTA's own pipeline first and since deployed for client use. The six other entries in the system log are marked scoped, which means the architecture, sequence and stack are worked out but they have not been built for anyone yet. The studio is new and would rather say that than dress up seven case studies it can't stand behind.

What does the Lead Intelligence Agent do?

It pulls accounts from Apollo, enriches them against firmographic and hiring signals, scores fit with a Claude-driven rubric tuned per ICP, writes a first-touch email in the founder's own voice, and pushes the sequence to Lemlist with replies synced back to HubSpot. One operator reviews before anything goes out; nothing sends unreviewed. It runs on Claude API, Next.js, Node, Postgres and Vercel.

What does “scoped” mean in the system log?

Scoped means designed but not delivered: the inputs, the sequence of steps, the failure points and the stack are all worked out, and none of it has been built for a paying client. It is an architecture ANTA is ready to build, not a claim about work already done. Anything that has actually shipped is labelled shipped and named.

Where does the 11.4% reply rate come from?

From ANTA's own outbound, run through the agent across roughly 2,400 runs and a few hundred scored accounts. Treat it as one sender's number against one ICP, not an industry benchmark — the honest teardown of what worked and what broke, including three named failure modes, is in the note on scoring B2B leads with an LLM.

Can you build the same system for us?

Yes, and it gets rebuilt against your ICP, your data and your CRM rather than cloned. The scoring rubric is the part that has to be yours: a rubric tuned for someone else's buyer is the reason most off-the-shelf lead scores are unusable. The repository ends up in your account like every other build.

Can we see it running before we commit?

Yes — ask for a walkthrough on the intro call and you'll see the real system score live accounts rather than a recorded demo. That call is also where the first scope gets sketched, so it doubles as the cheapest way to find out whether there's a pilot worth doing.

04 / Contact

You already know what to build.

Send the two-paragraph version. You'll get a real technical response, not a calendar link and a discovery questionnaire.