I build the systems
behind the product.
Five years on the UI, the APIs, and the jobs that have to keep running. Currently at Topo (YC W24). Before that, Artisan AI (YC W24) and LeadSquared. I usually own a feature all the way through, and most of it is still in production.
A few numbers
from production.
Daily volume on systems I still work on at Topo, plus deliverability at Artisan.
Selected work
still in production.
Three of them in detail, then a few more below.
HubSpot integration
Customers connect their HubSpot account to Topo. From there we sync leads, companies, and activity into HubSpot, and we react when something changes on the HubSpot side — like a deal closing — so Topo stops reaching out.
Customers run sales in HubSpot and outreach in Topo. They need activity from Topo to land in HubSpot, and if a deal closes or a company moves to opportunity in HubSpot, Topo has to stop reaching out.
Temporal workflows that push leads, companies, and activity into HubSpot. Webhooks the other way — when HubSpot says a deal closed or the lifecycle changed, we record a meeting in Topo and exclude that company from more outreach. The webhook path handles duplicates, retries, and bursts. Later I added owner sync so the same person owns the record on both sides.
Runs for 200+ customers, about 50–100K leads a day.
Daily sales workspace
A React workspace for email, calls, and LinkedIn, plus the API under it.
Reps were jumping between three tools, with no shared queue and no reliable writeback to the CRM. I built Tasks and the Daily Sales Workflow: assignment, reminders, bulk actions, CRM sync, and AI drafts in one place.
It got faster from pagination, indexes, query work, and caching. We didn't add more machines.
Half the email wasn't landing
Artisan is an outbound product. At 50% delivered, half the send volume was wasted.
Deliverability was at 50% when I joined. I reworked sending, quality checks, and filtering, and got it to 95%. The scheduler now holds 10K+ emails a day.
I put analytics in at the same time, so a drop shows up before a customer reports it.
Rebuilt subscriptions, invoices, upgrades, and cancellations on Stripe. The old path dropped state without anyone noticing.
Reply drafts and call icebreakers sit in the task list reps already use.
Automated production database upgrades. Downtime is two or three minutes.
Database access, approvals, alerts, and scheduled logs across clusters and regions.
The UI, the jobs,
and the infra.
I usually own all three. Scroll to step through each.
Five years,
three startups.
Paris, France · Remote
- Built the HubSpot integration for 200+ customers — 50–100K leads a day. Temporal jobs push leads, companies, and activity into HubSpot; webhooks pull deal and lifecycle changes back so we stop outreach; owners sync both ways.
- Built the daily sales workspace: email, calls, and LinkedIn in one queue. 10–20K tasks a day, with assignment, reminders, bulk actions, and CRM writeback.
- Built Temporal workflows for HubSpot dynamic lists — customers link a list, we import the leads, then pause, resume, detach, and recover when a run fails halfway.
- Scaled outbound to 20–40K emails a day, and added sequence analytics and sending-health so a bad day is visible.
- Made lead management faster with pagination, indexes, query work, and caching — not by adding machines.
- Embedded AI reply drafts and call icebreakers in that same queue, not in a chatbot on the side.
San Francisco, CA · Remote
- Took email deliverability from 50% to 95% — sending path, quality checks, and filtering. For an outbound product, that number is the product.
- Hardened the scheduler to hold 10K+ emails a day without falling over.
- Added analytics for lead and email activity, so a drop shows up before a customer emails us.
- Rebuilt the Stripe subscription path end to end. A billing bug doesn't show up as a ticket — it shows up as lost revenue.
- Shipped HubSpot and Salesforce integrations for lead sync and engagement.
Bangalore, India
- Automated production releases and regional infra provisioning, so deploys weren't a manual DevOps ritual.
- Built RDS upgrade workflows in production: downtime from a long outage down to two or three minutes.
- Built the internal DevOps portal — database access, approvals, notifications, and scheduled logs across clusters and regions.
- Sped up Advanced Search on high-traffic paths through caching and backend work.
What I use
day to day.

Hi, I'm Yatharth
I'm a full-stack engineer. Most of my work sits between the product people use and the systems that have to keep running — CRM integrations, background jobs, email that has to land, and the infra underneath.
I've done this at LeadSquared in Bangalore, Artisan AI in San Francisco, and Topo in Paris. At each one I've usually owned a feature end to end: the UI, the API, and the jobs behind it.
I'm based in India, open to senior or founding roles, and happy to talk if that sounds useful.
Lovely Professional University · CGPA 9.0/10 · 2017–2021
Redis internals
A side project,
and a few videos on internals.
Tricity Rent is live. The videos are me working through Redis and uv.
A rent map for the Tricity
A free, anonymous map of rents people actually pay in Chandigarh, Mohali, Panchkula, Zirakpur and Kharar. Medians by sector and BHK. Locations are fuzzed to about 100 m. No brokers, no login.
Résumé
on the page, and as a PDF.
Same content either way. Use whichever is easier.
yatharthverma.dev · GitHub · LinkedIn · YouTube (3.6K+)
- Designed and evolved CRM integration architecture for a platform serving 200+ customers and processing 50K–100K+ leads daily — bidirectional HubSpot sync with entity-level source-of-truth semantics, webhook-driven updates, conflict resolution, customisable mappings and lifecycle-safe reconciliation.
- Built the Tasks and Daily Sales Workflow: a unified multi-channel workspace processing 10K–20K tasks/day across email, calls and LinkedIn, with CRM sync, assignment pools, reminders, bulk actions and AI-generated reply drafts.
- Engineered reliable asynchronous workflows with Temporal for CRM-backed dynamic lists — reconciliation, pause/resume/detach lifecycles, external state changes and failure recovery.
- Scaled high-volume outbound to 20K–40K emails/day, building sequence analytics and sending-health diagnostics while improving lead-management performance through pagination, indexing, query optimisation and caching.
- Embedded AI into core sales workflows: AI-generated messaging, email drafts and lead-specific call assistance.
- Improved email deliverability from 50% to 95% by optimising sending, email quality and filtering systems.
- Engineered and hardened the email scheduler to reliably process 10K+ emails/day.
- Built analytics infrastructure for real-time tracking of lead and email activity.
- Redesigned the subscription and billing workflow with Stripe, rebuilding the end-to-end lifecycle.
- Built CRM integrations for HubSpot and Salesforce, enabling lead sync and engagement workflows.
- Automated the production release process and regional infrastructure provisioning, reducing manual DevOps effort.
- Engineered automated RDS upgrade workflows, reducing database upgrade downtime to 2–3 minutes.
- Built internal developer/DevOps tooling for controlled database access, approval workflows, notifications and scheduled log reporting across clusters and regions.
- Improved performance of high-traffic product workflows, optimising Advanced Search through caching and backend improvements.
Jalandhar, India · CGPA 9.0/10 · 2017–2021
What I'm looking for
next.
A few specifics so you can skip the email if it's not a fit.
If you're hiring,
email me.
If you're a founder and the product is growing faster than the systems under it, email me. Integrations, volume, reliability, in-product AI.


