Email0+mailboxes0delivered
0%reply rate
Sequence
- 1Intro
- 2Follow-ups
- Every inbox warmed up and reputation-scored; below the bar it sends less, automatically
- Warm-up ≥90% gates send volume, with daily limits enforced per inbox
9+ years running growth taught me where the bottleneck sits. Now I build the AI automation that clears yours.
A founder, a marketer and an engineer see three different problems in one funnel. I think in all three, so I find the one that matters. Outbound is one of them, and the system below runs in production.
Book a 30-minute teardown
You'll leave knowing what to automate first
Intent signal
Capture
≡
Screen
Profile Enrichment
Person Fit Check
Company Enrich
Company Score
@
Email Enrichment
SMTP
Email Infra
Write Email Copy
SmartLead
LinkedIn Infra
Write LinkedIn Copy
HeyReach
%
Analyze
↻
Learn
◆
Plan
Health Check
Self-healing · always on
- 0+
- qualified meetings booked / month
- 0
- outbound pipelines
- 0%
- email reply rate
- 0%
- LinkedIn reply rate
- 0%
- LinkedIn connection acceptance
The flagship in numbers, then the estate it feeds.
One fresh-intent pipeline books most of the meetings. Four more run underneath it on the same qualification, copy and delivery layer. Here's the flagship in depth first, then the whole estate.
Intent, in detail
The numbers below are the fresh-intent pipeline alone, for one cycle. The other pipelines run on the same qualification layer but their own signals and schedules.
Someone likes a post from a person our buyers follow. Thirteen stages later they are a qualified lead in a campaign with copy written for them. Nobody touches it.
Every stage that costs something sits behind a stage that costs nothing. A cheap screen on the job title we already have runs before any enrichment API is called. Whether the person fits runs before we enrich their company. The company score runs before we go looking for an email address. By the time we spend real money on a record, three separate filters have already agreed it is worth it.
It runs twice a day and drains each stage to empty before moving on, so an interruption costs time and not correctness.
Sourcing, which fans out
Accounts watched
0The seed profiles tracked for signal
Posts judged4.9×
0Each scored on whether its audience is worth mining
Engagers captured5.0×
0Likers and commenters, once engagement had settled
Qualifying, which collapses
Headline screened35% kept
0Deterministic filter, no API spend
Profile enriched72% kept
0Cache first, API only on a miss
Person fits60% kept
0LLM relevance score
Company enriched47% kept
0Profile plus site scrape
Company scored ≥551% kept
0Negative-selection scoring
Email found85% kept
0Three finders, then MX verify
Delivery, which splits by channel
Every message is written off the prospect's most recent LinkedIn post, their company and them, then tailored to the channel it goes out on.
EmailLinkedIn0+accounts0delivered
0%connection accepted
0%reply rate
Sequence
- 1Connection note
- 2Intro
- 3Follow-up 1
- 4Follow-up 2
- 5Follow-up 3
- No warm-up — safe daily connection and message limits per account
- Each account's profile is matched to the target's geography
- Campaigns are split by region so the sending profile always fits who it's reaching
Across all pipelines
- 0.0M
- Cached lookups
- Profile and company hits served without paying an API
- 0
- Campaigns run
- Across email and LinkedIn
- 0
- Pipeline runs recorded
- Every run logged, funnelled and reported on
Read once from production, July 2026. Totals only, nothing live.
Five ways in, one way through
Outbound isn't one pipeline. Five of them start from different signals and converge on the same qualification, copy and delivery layer, backed by the loop that keeps all of it running unattended.
Fresh Intent
FlagshipEngagement on posts from accounts we watch
Full rate — 15% email · 30% LinkedIn reply · 40% connection accept
- Discover
- Mine the signal
- Filter cheap to expensive
- Qualify the company
- Route, write, ship
Twice a day. The flagship, and the source of the funnel below.
Old Intent Data
Intent leads already sitting in the outreach platforms, never replied
~25K reached (15K email · 10K LinkedIn) · half the fresh-intent rate
- Extract from history
- Enrich and qualify
- Back into campaign
Weekly and deliberately off-peak, because it shares rate limiters with the daily runs. High volume, colder audience.
Email to LinkedIn
Leads whose email sequence ran out without a reply
~11K leads · 20% accepted · 15% reply (LinkedIn only)
- Fetch non-responders
- Dedup
- Re-check the person
- Re-check the company
- Ship on LinkedIn
Twice a week. A second touch on a different channel costs almost nothing, because the enrichment is already paid for.
AI Voice SDR
Curated lists, routed by region
~50-60 calls/day · ~5 meetings/day
- Split by geography
- Qualification gate
- Place the call
- Write the outcome back
Built in n8n on Retell rather than in the TypeScript stack. A voice channel alongside email and LinkedIn.
AI SDR — Funding Signal
Funding announcements pulled from the news
~6.5K reached (email + LinkedIn) · half the fresh-intent rate
- Ingest the signal
- Resolve to companies
- Qualify the company
- Find the right person
- Route, write, ship
Daily, on its own tenant. A company that just raised is a company that just started spending. Newest and smallest.
Then all of them share this
Score the companyWrite the copyDeliver
EmailLinkedInVoice
One scoring prompt, one set of copy frameworks, one delivery layer. A new pipeline only has to produce qualified leads; everything after that already exists.
What keeps it running
Five pipelines are only worth having if they stay up without anyone watching them. Three loops make that true, each on its own schedule.
The self-healing loop
It reads the last 24 hours, works out what broke and why, posts to Discord, opens GitHub issues, and ships approved fixes.
- Cron, every morning
- Read 24h
- Classify failure
- Seen before?
- Repair or file it
- Report
Email infrastructure
Cold email only works while the mailboxes stay healthy. This watches every one of them and pulls the burned ones out.
- Nightly sync
- Mailbox
- Warm-up score
- Health grade
- Rotate or pull
- Send
The operations console
Live row counts across the whole database, conversion at every stage, and campaign performance by channel. When a number moves the wrong way I can see which stage did it.
- On load
- Postgres
- Aggregate
- Funnel + runs
- Console
What it's built from
Orchestration
0cloud tasks, 14 crons
Trigger.dev v4, not a no-code tool like n8n. Every stage drains its own queue to empty before the next one starts, and selects its work with a query rather than being handed a batch, so a crash mid-run costs time and not correctness.
Data layer
0tables, 2.5M cached lookups
PostgreSQL with Drizzle. Cache-first on a 60-day TTL, which is what keeps enrichment spend flat while volume grows. A repeat profile costs nothing.
LLM layer
0typed functions
Claude and GPT behind BAML, so every model call returns a typed structure instead of prose to parse. Scoring is negative-selection: the model argues why a company fails, which is far more reliable than asking it to rate one.
Email waterfall
0finders, then MX verify
The only genuine waterfall in the system. Providers are tried in cost order and stop at the first hit, then every address is MX-verified before it is allowed anywhere near a campaign.
Delivery and infra
≥0%warm-up gates send volume
Managed mailboxes scored on warm-up, bounce and blacklist health. An inbox that drops below the bar sends less, automatically, before it can take a domain down with it. Replies attribute back through HubSpot.
Self-healing
0safety rails on every auto-fix
Stall watchdogs on every stage, reporting into Discord. Known failures repair themselves from a whitelist of pre-written SQL behind a row cap, a cooldown and a dry-run default. No model writes a query that touches the database.
What would this look like on your pipeline?
Most of what's above started as somebody's manual Tuesday. If you can describe the version of that you're living with, I can usually tell you which part is worth automating first.
What I build with
- Build
- TypeScript, Node.js, Python, PostgreSQL, REST APIs, Bash, GitHub Actions
- AI systems
- Claude & Claude Code, Codex, GPT, Gemini, BAML, Prompt engineering, Agentic workflows, Custom plugins & skills
- Orchestration
- Trigger.dev, n8n, Retell AI, Cron scheduling, Queue draining, Observability
- GTM infrastructure
- Outbound pipelines, Lead enrichment, ICP filtering, Email deliverability, CRM automation, HubSpot, SmartLead, HeyReach
- Growth
- Google Ads, Meta, LinkedIn, Amazon Ads, CRO, Budget pacing, ROAS & CPL, Lead scoring, GA4, GTM, Looker Studio
Find them, convert them, run it.
Almost every growth problem is one of three things: not enough of the right people arriving, too few of them converting once they do, or too much of the business still being done by hand. The method doesn't change between them.
Systems that find customers
The expensive part of demand generation isn't the tools, it's the manual loop running between them. These replace the loop.
Outbound pipelines (built or run professionally)
Signal to qualified lead to written copy to delivery, on a schedule, with nobody in the middle. The engine above is one of them.
AI voice SDR (built or run professionally)
Calls, qualifies, routes and writes back to the CRM without anyone listening in. Region-aware, with its own daily caps.
Email infrastructure (built or run professionally)
Warm-up, health grading and rotation across hundreds of mailboxes, so deliverability stops being luck.
Systems that convert them
Traffic that doesn't convert is just cost. This is the half I spent nine years and up to €1.5M a month learning before I started building.
Paid media (built or run professionally)
Google, Meta and LinkedIn at scale, with the pacing and account structure that holds a target return rather than chasing it.
CRO and attribution (built or run professionally)
Landing pages, audience testing and A/B, with the GA4, GTM and HubSpot tracking wired underneath so a test means something and a booked call traces back to the campaign that caused it rather than being argued about later. Part of how the 2 to 3x month-on-month at Social Beat happened.
Shopify (built or run professionally)
Storefront build and optimisation, then the funnel around it.
Systems that run the rest
Most of the manual work inside a company was never customer-facing. It's the spreadsheet nobody owns, the handover that gets forgotten, the report someone rebuilds every Monday.
Custom AI automation (built or run professionally)
Any repeated manual process turned into something that runs on a schedule, reports on itself and says so when it breaks. That's what everything above this section is.
Custom apps and design systems (built or run professionally)
Bespoke internal tools and interfaces for when the off-the-shelf option doesn't fit the process — built on a real token system rather than a template, so the tenth screen still looks like the first. The Mhari console and this page are both these.
Dashboards and data analytics (built or run professionally)
Live numbers read from the source instead of a monthly export, plus the tracking underneath that makes them trustworthy. The operations console further up this page is one of these.
Both tracks get built the same way: find the expensive manual loop, then build the thing that runs it and watches itself.
TypeScriptTrigger.devPostgreSQLRAG and agentic workflowsTyped LLM contractsSelf-healing watchdogs
Nine years in paid media, then I started building.
Up to €1.5M a month at GroupM. Running growth at that scale shows you which decisions actually need a person, and which ones are coordination a system does better.
- 2017 — 20200+ leadsIndia · UAETrackingOptimizeControlAutomate
Cbetter Consultancy
SEO & Media Buying Specialist
The fundamentals, learned end to end: run the ad, track it, improve the page.
Brands
- 2021 — 2024₹0.0Cr / moIndia · 2–3× MoM growthTrackingOptimizeControlAutomate
Social Beat
Lead, Digital Media
The full stack under one roof: paid, commerce, tracking, dashboards, CRO.
Brands
- 2024 — 2025€0.0M / moEurope · APACTrackingOptimizeControlAutomate
WPP · GroupM Nexus Global
Manager, Paid Media
Scale with discipline: hold a consistent ROAS at €1.5M a month, watch everything, never let performance slip.
Brands
- 2025 — now0 pipelinesAI AutomationTrackingOptimizeControlAutomate
Qubit Capital
AI Growth Systems Engineer
Nine years of judgment, turned into systems that run themselves. The manual monitoring became self-healing.
Four acquisition pipelines in production. Outbound that runs without a person watching it. Nine years of judgment, encoded as systems
Other things I've built.
Some shipped and in use, some in progress, some scaffolded with the research done and the build not started. Labelled honestly, newest proof first.
Shipped
Running and in useOpen live
Your Flow State
Passion project · Sonic Weaver
A browser visual engine that listens to sound. Low, mid and high frequencies each move a different part of the picture, and the whole thing stays one image instead of three.
JavaScript · WebGL · Web Audio · Storybook
Paalthi
Built for a café
Expense and sales tracking for a café, on Supabase and Vercel. No build step, plain HTML and JS, with three small serverless functions for the one thing that needs elevated privileges.
Supabase · Vercel · Vanilla JS
Mhari Ops Console
Client internal tool
Customer feedback capture and an admin dashboard for a jewellery brand, backed by Shopify and running on Vercel serverless. Tracks which review links went out, which got opened, and which turned into an actual rating.
JavaScript · Shopify · Vercel
Outbound Builder
Claude Code plugin
A plugin that packages everything I learned building the Intent Engine into something any Claude Code session can use: 24 implementation docs, 25 code templates, 8 skills and an architecture agent.
TypeScript · Claude Code
CodeHealth Check
Claude Code plugin
The self-healing loop, pulled out of Qubit's codebase and made generic. Collects health signals from any system, scores it 0-100, fingerprints and dedupes issues, discusses them in a chat channel, and on approval opens GitHub issues and runs safety-gated repairs.
TypeScript · PostgreSQL · Discord · GitHub
CodeBuilding
In progress right nowTrade for All
Side product
Export compliance and tariff research for small manufacturers.
Performance Marketing System
In progress
The paid media half of the same idea. Nine years of campaign judgement, encoded as a system that runs accounts rather than reports on them.
Idea
Scaffolded or researched, not builtAll-in-One DJ
Visual prototype
One tool for the whole DJ prep workflow instead of four that don't talk to each other. Currently at the stage of working out how it should look and behave.
The other half
I perform and make things under the name Your Flow State. Electronic music, DJing, abstract photography, calisthenics.
This is not a hobbies section. The way I build is the way I mix: find the rhythm, iterate in small passes, keep the whole thing coherent instead of impressive in pieces, and show up consistently over a long stretch. The audio-reactive engine exists because those two halves are the same brain, and I wanted to see what happened if I pointed the engineering at the music.
- Electronic music & DJing
- Performing as Your Flow State
- Abstract photography
- Composition practice, mostly light and motion
- Calisthenics
- The long-term consistency argument, applied to a body
- Audio-visual systems
- Where the two halves finally met
Tell me what you're building.
I take on a small number of consulting projects at a time.
Tell me roughly what your outbound looks like now and where it's costing you time. I'll come back with what I'd change first, whether or not it turns into work.
Book a 30-minute teardown
Bring how your outbound runs today. You'll leave knowing which stage is leaking and what I'd fix first, whether or not we end up working together.
- Based
- India, Remote
- Working
- Consulting and freelance
- Studied
- B.Tech, Information Technology JECRC University, Jaipur, 2013 — 2017Digital Guru Expert (Black Belt) Google
