Pieces full-timeOpenHands contractOpenObserve contractSuprSend first marketer
Stack: Python · TypeScript · SQL (BigQuery, Postgres) · Apache Beam · GA4 & GTM · HubSpot · RAG and agents · Next.js on GCP and Vercel.
What I do
The systems that do the work nobody has time for.
Most teams don't need more tools. Or more headcount. They need the plumbing between what they already have. And someone who builds it, ships it, then leaves it running.
Pipeline
Outbound that finds real buyers
Find who's actually using your product. Or shopping your competitor right now. Enriched, drafted, dropped in your reps' inbox every morning.
Shipped: the Stark Agent + Iris copilot at OpenHands · a personalized demo engine at OpenObserve
Clarity
Attribution you can trust
One source of truth across your scattered tools. Finally answer which channels bring customers who stick. Not just who sign up.
None of that client work started from a blank page. It started here. I owned marketing data and marketing AI at Pieces (Series A, $13.5M) and built five systems the growth team kept needing, because no tool on the market solved them. Here's what each does for the business. The engineering is one click away.
Built at Pieces
Your team's memory
Ask what did we decide, what shipped last week, what are customers complaining about. Get a sourced answer in seconds. No more 15-minute hunt through chat, docs, and meeting notes.
min to find one answer15
→
seconds2.8
How it works (technical)
A retrieval system (RAG) over 50+ chat spaces, daily transcripts, GitHub, and product docs. Runs on Google Cloud. The hard part isn't search. It's relevance. Every retrieved doc gets graded before it's used, so answers don't mix last quarter's strategy with this quarter's pivot. Same knowledge base feeds the outbound and content systems below.
$ask "What did the growth team decide about the Dev.to campaign?"
The team decided to double down on Dev.to.
Attribution showed 34% activation vs 12% from paid.Sources: Growth Sync (Jan 21), Campaign Review (Jan 18)
Answer in: 2.8s (vs ~15 min of digging)
Finds the conversations worth joining. The dev quietly evaluating you. The comparison thread. The buying-intent question. Scored and dropped in Slack. The noise gets filtered out.
found by hand / day0–2
→
real ones / run12
How it works (technical)
A single tweet means nothing alone. “Just tried the copilot” could be praise or a complaint. The system walks the full reply chain, enriches the profiles involved, then scores the whole context with AI. Runs on a schedule from the cloud without getting IP-banned. Backs off when rate-limited. Falls back to sequential if parallel fails. No manual restarts.
Answers the Monday question every growth lead has. Which channels bring customers who actually stick. Not just who sign up. One source of truth, pulled from tools that never talked to each other.
disconnected tools5
→
query for the answer1
How it works (technical)
The whole measurement stack, built from scratch. GTM tag management and event taxonomy. Apache Beam pipelines pulling web, social, and product-signup data into BigQuery. A lifecycle model joining session to feature usage to retention, with multi-touch attribution. Pipelines are idempotent, so a failed run retries without duplicates. New client means new config, same code.
source │ signups │ D7 retention
─────────────────┼─────────┼──────────────
dev_to_blog │ 342 │ 34%twitter_organic │ 187 │ 28%
linkedin_ads │ 523 │ 14%
One query across 5 tools that never talked.
The channel with the most signups had the worst retention.
Product-qualified leads, not stargazers. The people actually running your product in production. Or shopping your competitor today. Each one researched, scored, and handed to your reps as a draft. They review and send.
min research / lead15
→
min, evidence attached5
How it works (technical)
Signals from 9 platforms: GitHub, Docker, PyPI, Stack Overflow, job boards, competitor comparisons, and more. Linked into one identity per person. Scored for real production evidence: a config committed, a CI/CD workflow, an org-owned image. Replaces 3 to 4 tools (Apollo, Clearbit, reo.dev, Sales Navigator). Refuses to draft for low-confidence matches, so reps never spam someone who just starred a repo.
Lead:karan-sharma · Zerodha (India's largest broker)Verdict:Active production user · high confidence
Why this lead is real:
├── Running the product in CI/CD
├── 11 commits in 2 months
├── Public project built on top (18★)
└── Company confirmed via GitHub bio
→ Draft written. Rep just approves.
One blog post becomes native posts for Dev.to, LinkedIn, X, and Medium. Written to each platform's rules. Quality-scored. Deduped. Seconds, not the 2+ hours it takes by hand.
hours of repurposing2
→
seconds12
How it works (technical)
Every platform has different rules. Dev.to wants code depth. LinkedIn wants 1,300 professional characters. X wants a 280-character hook. The system scores each variant on a 100-point rubric and refuses to publish below 70. If two variants are more than 85% similar, it regenerates on its own. No daily babysitting.
Five systems, five separate builds. The pattern underneath them was always the same: pull from the tools that never talked, judge what came back, and refuse to answer when the evidence is thin. Project Cortex is that pattern built once, deliberately, with no client attached.
The first hire is Atlas, a data analyst. You ask why signups fell, in plain English. It looks across GA4, HubSpot, your warehouse, GitHub and Slack, throws out the explanations that only look right, and tells you what happened, what to do, and how much to trust it. Any sentence it cannot back is deleted before you see the report. The second hire is a config file, not another product.
The best thing it has done is refuse to answer. Pointed at a live CRM showing revenue up 7x, it found five of the wins shared one import timestamp, checked the open pipeline unprompted, and recommended auditing the close dates before anyone reported the number. A dashboard would have put that 7x in a board deck.
Everything above is my account of it. Here is theirs.
“During Nikhil's tenure at Pieces, he led both marketing data infrastructure and marketing AI systems. I recommend Nikhil to teams building developer-focused products that need someone who can architect and execute technical GTM systems end-to-end.”
Tsavo Knott, CEO & Technical Co-Founder, Pieces (Series A, $13.5M)
LinkedIn Recommendations
Tsavo Knott
CEO & Technical Co-Founder, Pieces for Developers
February 2026
“During Nikhil's tenure at Pieces, he led both marketing data infrastructure and marketing AI systems. He built Osiris, a marketing RAG system on GCP that indexed internal docs, GitHub content, and internal conversations into a queryable assistant. On top of that, he developed custom agents for social listening, content workflows, and outbound personalization. He also owned core data plumbing end-to-end: GA4 integrations, GTM setup, and Apache Beam -> BigQuery pipelines that consolidated web, social, and product-signup signals into a usable GTM data layer. Nikhil worked effectively across engineering, data, and growth partners. He was consistently strong at connecting GTM goals to technical implementation and delivering systems from design through production. I recommend Nikhil to teams building developer-focused products that need someone who can architect and execute technical GTM systems end-to-end.”
How I got here
I started in marketing. Then got tired of waiting on systems that didn't exist.
Before Pieces I was the first marketer at SuprSend. New category, infra startup. No marketing ops. No RevOps. No data team. The campaigns worked. But every workflow I needed, I had to build myself.
That's when it clicked. The campaigns were never the bottleneck. The systems were. So I became someone who builds them. We didn't just rank on G2. We created the notification-infrastructure category and took #1.
