Kritwish Mondal · Senior GTM Engineer
Clean data. Real signals. Measured AI.
I build the systems that keep revenue data true, turn signals into pipeline, and put AI into production only after it has been measured. Every number on this site comes with its receipt.
Example run log: run account-matcher --backlog. 1. Audit: 7,175 leads to attach to accounts 2. Snapshot: before-file written for rollback 3. Pilot: 200 leads matched, OK 4. Apply: 72/72 batches OK 5. Verify: accounts 1,192 → 4,136 · 0 duplicate domains ✓
- new ARR in one month from automated outbound, 42% of that month's total
- $104K+
- new ARR in one month from automated outbound, 42% of that month's total
- cold-call pickup rate, 15% → 45%, with AI voicemail detection
- 3×
- cold-call pickup rate, 15% → 45%, with AI voicemail detection
- meetings booked in the best month
- 200+
- meetings booked in the best month
- leads converted onto the right accounts after the CRM rebuild
- 106 → 4,800
- leads converted onto the right accounts after the CRM rebuild
Builds with
- Salesforce
- HubSpot
- Gong
- lemlist
- Instantly
- Clay
- FullEnrich
- ZeroBounce
- Leadfeeder
- Slack
- Claude / Anthropic
- OpenAI
- LangGraph
- Langfuse
- n8n
- Python
- TypeScript
- Postgres
- Redis
- DuckDB
- Next.js
- Vercel
- Google Cloud Run
- Metabase
- Stripe
- Chargebee
- Twilio
01 / The work
Three pillars.
Everything I've built falls into one of three buckets. Each has its own page with the systems, what they did, and how it was checked.
02 / Systems
Everything I've shipped.
01 · Clean Data
- Salesforce rebuildLead-status sync 45% → 99.99%; ~4,800 leads converted, up from 106Shipped
- Lead and account hygiene7,175 leads matched to accounts with 0 duplicate domains; names, sources, emails and titles cleaned automaticallyProduction
- Drift monitorA nightly full-history copy of Salesforce, read by 56 SQL checks with a Monday alertProduction
- Suppression engine5,951 records suppressed, 42/42 end-to-end tests, and a sync-loop incident fully reversedProduction
02 · Pipeline
- Inbound lead deskQualify → route by state → Slack alert → Day 0/2/4 cadence; 101/101 on a real-data replayProduction
- Signal tracking and hot accountsSignals, research and enrichment roll up into one heat score; the hottest accounts and their buyers land with BD every MondayProduction
- Lead enrichmentOwner first name found for 86% of local businesses, at ~$0.012 a leadProduction
- Campaign engineFour campaign types on one runner. AI writes, code decides, and copy that keeps failing the spam check is droppedProduction
- SDR lead scoring and routingEvery open, click and reply scored in real time and routed to an SDR who's in todayProduction
- AI voicemail detection for cold callingPickup rate 15–20% → 45–50%: SDRs are only routed leads with a high chance of being picked upProduction
03 · Measured AI
- Call intelligence with a blind judge0 hallucinated values in 1,898 extracted entities; model picked by a blind LLM judgeProduction
- Knowledge layer with eval gatesRecall@5 100% and 0% false-confident answers on the eval set; evals block mergesProduction
- Marketing ledgerUnlinked opportunities: 49 of 105 → 1 of 104; attribution that matches the analysis sheet row for rowProduction
03 / Playbook
The same loop, every time.
Every GTM system I put into production goes through the same six steps: CRM changes, data hygiene, workflows and AI agents alike. That's why the numbers above have receipts.
- 01
Measure
Start with a baseline: what's broken, how often, and what it costs. Every system here has a before number.
- 02
Plan
Write down the rules, the edge cases and what done looks like, and get them reviewed before any code.
- 03
Build safely
Sandbox first, with tests and evals on every change. Anything customer-facing gets a human gate and a kill switch.
- 04
Roll out
A dry run that writes nothing, then a small pilot, then everyone. Every bulk data change keeps a rollback file.
- 05
Verify
Read the result back from the live system, whether CRM records, sync status or model output, and compare it with the plan. Trace every AI call and its cost.
- 06
Guard
Data-quality checks, monitors and alerts, so drift or a regression shows up within days and someone is told.
04 / Career
Where I built it.
Aug 2026 – Present
Senior GTM Engineer · CertifyOS
Healthcare provider credentialing and data platform
- Built the GTM engineering function from zero: the revenue stack (Salesforce, HubSpot, Gong, lemlist) and the automations and AI agents behind prospecting, routing, enrichment, call intelligence and reporting.
- Rebuilt the CRM until pipeline data could be reported on: lead-status sync 45% → 99.99%, about 4,800 leads converted onto the right accounts, and nightly data-quality checks.
- Shipped an account-heat engine that sends BD its hottest accounts every Monday, and AI that is measured before it's trusted: a call-intelligence agent picked by a blind LLM judge, and a knowledge agent with 100% recall@5 on its test set.
Nov 2021 – Aug 2026
Founding Member, GTM Engineering & RevOps · Zoca
AI-powered growth platform for US beauty and wellness businesses
- Built GTM engineering from scratch: hired a 4-person team (3 engineers, 1 analyst) and stood up an 8-person SDR team.
- Automated outbound end to end, from AI lead research and AI-written campaigns to reply scoring and SDR routing. It drove $104K of $250K new ARR in one month.
- Added AI voicemail detection to cold calling (pickup rate 15% → 45%), and predicted churn from customer conversations at 85% monthly accuracy.
May 2021 – Oct 2021
Data Scientist · HealthifyMe
Health-tech platform for coaching and nutrition
- Analysed user behaviour and engagement with SQL and Mixpanel to guide product decisions.
- Built a coach-scoring model that predicted coach performance across cohorts. Allocating by it lifted plan renewals 22%, cut churn 3% and contributed to a 12% ARR increase.