GTM Engineering

Find the right accounts.
Give sales a reason to reach out.

I build signal-based prospecting, account research and CRM workflows for B2B teams—connecting buying signals to relevant outreach and measurable follow-up.

Discuss your GTM system Explore selected work

Available for remote contract engagements.

Selected GTM systems

A closer look at the work.

Three systems for account research, reliable CRM operations and revenue analytics. Explore what they do, how they were tested and where their limits are.

RevOps Sync Control Plane

CRM sync.
With a recovery plan.

Keep account identity, field ownership and delivery order explicit—and make uncertain writes recoverable.

5-minute walkthrough → Inspect code →
Architecture, testing & limitations

Python · SQLAlchemy · HubSpot / Salesforce · Recovery & audit

Problem
CRM sync can duplicate accounts, overwrite good data or restore stale consent.
Implementation
Identity and field-ownership rules, version checks, fenced delivery claims, ordered outbox streams, read-back recovery and audited hold release.
Inspectable proof
5 checked demo scenarios; 123 tests passed, 1 skipped in the local 13 September 2026 SQLite run. Simulated trace: create → read-back → update.
Limit
No live CRM calls in the walkthrough. Concurrent ingestion and signed webhooks remain unimplemented; no revenue result is claimed.

Claude GTM Revenue Agent

Account research.
Grounded in evidence.

Turn public signals into account intelligence, with checks for unsupported claims and accounting for rejected-response costs.

Evidence console → Inspect code →
Architecture, testing & limitations

Python · Claude API adapter · FastAPI · Grounding & cost controls

Problem
Account research becomes a liability when AI invents facts or hides rejected-response costs.
Implementation
Public-signal ingestion, deterministic scoring, strict tool/output contracts, grounding checks, human-review boundaries and failed-response usage accounting.
Inspectable proof
12/12 deliberate corruptions rejected; 29 local tests passed in the 12 September 2026 receipt. This measures guard behavior, not live-model quality.
Limit
Briefs are offline mocks. No authorized live-Claude receipt, independent human benchmark or business outcome is claimed.

GTM Warehouse

Revenue analytics.
Built on tested models.

Bring account signals, touches and opportunities into a consistent SQL model that can be inspected and tested.

Warehouse walkthrough → Inspect code →
Architecture, testing & limitations

SQL · dbt Core · DuckDB · BigQuery · Dimensional models

Problem
Signals, touches and opportunities need a tested data model before funnel reports can be trusted.
Implementation
Staging and account/fact models, incremental signal modeling, generic and custom data tests, and fixes verified across two SQL engines.
Inspectable proof
35/35 data tests passed on real BigQuery on 27 August 2026, also verified on DuckDB. Build-node counts are not presented as test counts.
Limit
Signals are public-source data; touches and opportunities are synthetic. One batch does not prove recurring loads, production scale or commercial performance.

Discuss your GTM system → Explore supporting engineering work

Commercial impact

Experience

Employer scope and reported outcomes, kept separate from the public engineering demonstrations below.

Sydon · GTM Engineer 25,000+ → 150+ → 35+ prospects → qualified meetings → opportunities
MAPOG · First outbound hire 20,000+ → 600+ targeted prospects → qualified product demos across ~12 months

2026

GTM Engineer — Sydon.ai

Scope: Revenue Systems · GTM Strategy · Team Leadership. Built and operated AI-native GTM systems supporting 25,000+ prospects. Signal-driven research, qualification and activation workflows supported 150+ qualified meetings and 35+ opportunities — approximately 23% meeting-to-opportunity conversion across the tracked cohort. Reduced typical account preparation from 15–20 minutes to under 5 minutes, built 10+ operational workflows, shaped ICP and GTM plays, and led day-to-day execution across a small commercial team.

2025 – 2026

GTM Engineer / First Outbound Hire — MAPOG (GIS SaaS)

Scope: GTM Systems · Outbound Strategy · Team Leadership. As the first outbound hire, helped build the motion and supporting infrastructure from zero. Processed 20,000+ targeted prospects, generated and supported 600+ qualified product demos across ~12 months, and built 8+ automated workflows across Clay, Apollo, Bright Data, n8n and CRM infrastructure. Defined ICP, segmentation, qualification rules and campaign plays while leading execution across a small GTM team.

2022 – 2025

Co-Founder — Hahaha Marketing (digital agency)

Co-founded and operated a client-facing marketing business, owning customer acquisition, positioning, campaign strategy, delivery and commercial decisions.

Operating rigor

What changed after the systems met reality

The strongest proof is not a green demo. It is finding the failure, understanding it, fixing it and leaving a receipt.

Salesforce: made asynchronous behavior testable, not merely green.

The first real Apex run returned 19/20 because scheduled and batch behavior raced under Test.stopTest(). I isolated both paths into deterministic tests; the re-run finished at 20/20 with 96% org-wide coverage.

Inspect the test evidence

n8n: proved the fallback worked, then fixed the classifier.

A forced failure generated a page-intent payload through the conservative default, but n8n rewrote the error name and the code defect landed as “unclassified.” I updated the rule to the runtime payload the workflow actually received. External notification delivery is not part of this build.

Inspect the forced-failure receipt

Removed a stale metric from contact-facing output.

A live-data demo still pasted a hard-coded +183% into outreach. I moved the metric to the same dated source-of-truth function used by scoring, so the claim now recomputes with each run.

Inspect the correction

Made the documented setup genuinely reproducible.

The CRM lifecycle repository used Python 3.10+ syntax without declaring that minimum. I fixed the compatibility and setup documentation, then verified that regenerated outputs were byte-identical.

Inspect the reproducibility fix

Added uncertainty and sample concentration to my own metric.

Groundtruth's 60.5% precision was correct, but its confidence interval and 65% single-repository concentration changed how the result should be read. I published both caveats in the evaluation harness and ran the checker across this portfolio.

Inspect the self-audit
Role fit

What I can own end to end

From ICP and signal design to CRM activation, revenue data, observability and team adoption.

GTM Systems & Strategy

  • ICP development & segmentation
  • Signal design & account prioritization
  • Outbound plays & experiments
  • Funnel & pipeline analytics

Software & Data

  • Python / FastAPI / APIs
  • SQL / dbt Core / DuckDB / BigQuery
  • Postgres / Alembic / pgvector
  • Docker / Terraform / GitHub Actions

AI & Automation

  • Claude API / agent workflows
  • n8n orchestration & failure handling
  • Clay enrichment & qualification
  • Evaluation / grounding / human review

CRM & Operations

  • Salesforce Flow / Apex / SOQL
  • HubSpot ↔ Salesforce sync controls
  • Lifecycle / routing / governance
  • Playbooks / documentation / team adoption
Engineering evidence

Additional engineering work

Supporting implementations in evaluation, orchestration, Salesforce, enrichment and signal intelligence. Explore these after the three selected systems; each retains its own execution and data boundaries.

Explore additional systems, grouped by capability

AI / agent engineering & evaluation

Agents · evaluation · guardrails
GTM Eval Harness

Python, stdlib only · Wilson intervals · Precision/recall/F1 · Cohen's kappa

Tests whether GTM and AI claims are actually supported: confidence intervals, undefined-metric handling, subgroup concentration and README evidence checks. I used it to audit this portfolio, including my own published metrics.

87 tests passing · README-grounding checker run across 7 repos in this portfolio, including its own audit below · deterministic, no API key, $0
Live write-up → Code →
Groundtruth

Python · GitHub API · Defect scanning · Self-audited

Scans public repositories for a real defect pattern and reports false positives instead of hiding them. A second audit exposed uncertainty and sample-concentration caveats; both are now published with the result.

26 confirmed / 17 false positives → 60.5% precision · 95% CI [45.6%, 73.6%] · 65% of the sample is one repository — both caveats now stated openly
Live write-up → Code →
Revenue Brain

Claude Code workspace · Agent design · Logistic regression · openFDA API · Zero dependencies

Closes the loop from signal → brief → outcome → model update. Its learning logic recovers a hidden simulated response model; that validates the code path, not real-world reply lift.

1.51× lift vs random control after convergence · weight-recovery L1 error 0.600 → 0.098 · 1,000 real FDA records, 416 firms · outcomes simulated and labelled as such
Live write-up → Code →
Briefing Account Brief Generator Converts public sources into a traceable pre-call brief, reducing research without inventing contacts or unsupported facts. Every claim link-traceable to its source · no fabricated contacts · no email sent

Salesforce

Developer Edition deployment
Salesforce GTM Signal Layer

Salesforce · Apex (trigger + service + batch) · Flow · Validation rules · SOQL

Routes and prioritizes leads from live signals inside Salesforce using a custom signal object, before-save Flow, validation safeguards, Apex rollups and scheduled decay. Deployed and verified in a Developer Edition org.

20/20 Apex tests passing · 96% org-wide coverage · 200-record bulk transaction test · first real run was 19/20; a nested-async race was root-caused and split into deterministic tests
Live write-up → Code →

n8n

Live orchestration + failure handling
n8n GTM Orchestration

n8n · Scheduled workflows · Dedicated error handling · openFDA API

Turns public enforcement data into a ranked, grounded outreach queue on a schedule, with fixed-delay retries, rate limits and a separately exercised error path.

100 records → 34 firms → 11 cleared ≥60 → 10 briefs · error handler fired 80ms after a forced failure and generated a page-intent payload · no external notification delivery is claimed; the first classifier rule fell to “unclassified,” then was fixed
Live write-up → Code →

Signal & intent

Multi-source · source-grounded
GTM Signal Engine

Python · SEC EDGAR · GitHub · Hacker News · Retry-hardened

Combines real SEC, GitHub and Hacker News signals into an explainable Severity × Confidence × Recency score. Retry logic was added after two real live-run failures, then verified on a clean third run.

456 real signals, 14/14 entities, zero errors on the third run · not a prospect list — chosen for verifiable multi-source coverage
Live write-up → Code →
Atlas Signal Engine

Python · openFDA API · Signal scoring · Zero dependencies

Converts real FDA enforcement actions into signal-ranked account briefs grounded only in the underlying filings.

1,000 real openFDA enforcement records → firm-level scoring → top-8 briefs · $0 API cost · no fabricated contacts
Live queue → Code →
Signal Commit-Velocity Signal Engine Surfaces engineering-velocity inflections from public commit history as a dated, recomputed buying signal rather than a hard-coded outreach claim. +183% commit velocity on the dated write-up run (live data moves — it read +120% on 2026-08-23)

Next conversation

Let’s build your
next GTM system.

Need a GTM system built? Let’s discuss your targeting, research or CRM workflow—and what a focused engagement could deliver.

Available for GTM Engineering contract engagements.
Mumbai, India · Global / remote or contract.

Discuss your GTM system View résumé