Sium Akhter

Sium Ibn Akhter in a dark blazer and sweater, smiling

RevOps and CRM engineer. I fix the Salesforce, HubSpot, Marketo and Pardot setups revenue teams have stopped trusting.

Based in Dhaka, working US, UK, Canada and Australia hours.

Send me the symptomSee the work
Menu

I go and look

One live revenue system, 2026.

12,000 records and years of additive automation. Every check had passed. This is what I found under the dashboard.

Read the full case

  1. The dashboard. All 12,000 accounts scored, average 43, model marked healthy.
  2. The records under it. 1,789 populated accounts were scoring zero.
  3. The cause. A duplicate field was silently zeroing the score. It had cleared every standing review for months.

How I work

I’m a computer science graduate who ended up in revenue operations because that is where the interesting failures live. Across 300+ projects for 50+ clients in the US, UK, Canada and Australia, the pattern repeats: the dashboard is confident, the data underneath it is not, and nobody has looked in months. I go and look. Then I rebuild the process so the same failure cannot happen twice.

Every build ships with a runbook like this one, so the system works without me in the room. Sample content.

Four rebuilds

Client names are withheld on case studies, and the screens are redrawn with sample records. The full client list is further down.

2026

Buying group scoring engine

Rebuilt from zero

The existing model scored every account the same. Phase one did not find it partially broken. It found it non-functional.

Salesforce and Pardot, for a US enterprise print and marketing services company.

  • Audited the live model and established there was no usable signal in it before a line of new logic was written
  • Designed a three-layer capped architecture: intent and fit capped at 25, engagement at 40, buying motion at 40, with a velocity bonus of 10
  • Built the field foundation across Lead, Contact and Account with declarative rollups, so group-level scores aggregate without code
  • Shipped a record-triggered scoring engine on formula resources, reusable across objects
  • Delivered inside a 12-month engagement roadmap, executed to plan
2025–2026

Attribution and reconciliation flows

Two silent bugs closed

A submission counter nobody trusted, and a notification firing on merges it had no business firing on.

Salesforce and Pardot, for capital markets trading and risk software.

  • Reconciled submission counts across the marketing platform and the CRM, tracing the inflation to an automation rule re-sweeping historical records
  • Diagnosed a false notification triggered by Lead to Contact merges and separated merge traffic from genuine activity
  • Built an opportunity contact role attribution flow with an overwrite guard, so later touches never clobber first-touch credit
  • Added a scheduled-path retry for records arriving out of order, then verified the immediate path against live data
  • Designed a companion flow linking the original source contact to the opportunity, delivered against a formal client acceptance pack
2026

Pipeline rebuild and data governance

Migrated clean

A deal pipeline that had grown stages nobody could define, sitting on a property set nobody had pruned in years.

HubSpot, for US enterprise SaaS operations.

  • Rebuilt the sales pipeline with defined stage exit criteria, then migrated live deals against an agreed old-to-new mapping
  • Audited contact properties by real usage and fill rate, and retired every one carrying no signal
  • Standardized closed-won and closed-lost reasons as governed picklists, so win and loss reporting compares period over period
  • Cleared out the workflow layer left behind by the old pipeline
  • Produced the sales enablement material, so the new pipeline was adopted rather than worked around
2026

Automation debt teardown

83% fewer evaluations

Years of additive automation. Every rule reasonable on its own, the set of them collectively unpredictable.

Salesforce, for a live 12,000-record revenue system.

  • Found a duplicate field silently zeroing the score on 1,789 populated records
  • Corrected an archive query that had quietly excluded 1,720 accounts from every report built on top of it
  • Traced a data-loss path between the client’s web forms and the downstream platform to a browser-side redirect, and rebuilt the transfer as a server-side webhook
  • Consolidated six competing automated processes into one before-save process with six branches, removing execution-order risk across the lifecycle
  • Reduced active automation rules from 80 to 27 with no loss of coverage

Recognise one of these in your own system?

Send me the symptom

Projects delivered

300+

For 50+ clients in the US, UK, Canada and Australia since December 2024, as part of Get Levrg in Dhaka.

Tools I built

Built so nobody repeats the work. Each one replaced a manual job I kept seeing on client accounts.

  • WebSherlock

    2026

    An AI validator that resolves verified company domains from raw, unstructured input by cross-checking search results against LLM judgement. A 10-minute manual check now takes under 3 seconds, at 98%+ reliability, across datasets of 10,000+ records.

    Python, JavaScript, SERP API, LLM integration

  • InHunt

    2026

    An asynchronous enrichment pipeline that resolves raw CRM contact rows into verified LinkedIn identities in bulk, cutting manual lookup time by 90% and feeding clean data straight into outbound.

    Python, JavaScript, SERP API, Gradio

  • Salesforce UI optimization skill

    2026

    A repeatable audit-and-optimization process encoded as a reusable AI skill, so any team member runs the same quality of review without the underlying platform expertise.

    AI skill authoring, prompt engineering, process documentation

  • Smartling delivery tool

    2025

    Led execution and client communication on the engagement, then scripted a custom tool that reached 98% accuracy and compressed a three-month manual process into six days.

    Python, client delivery

  • Marketo lifecycle automation

    2026

    Multi-track nurture and lifecycle programs on trigger and batch campaigns with tokenized templates. Scoring and MQL entry rules mapped to matching CRM fields, so lifecycle stage reads the same in both systems. Documented as a runbook the team runs without me.

    Marketo programs, smart campaigns, CRM sync

  • Sync monitoring and data quality

    2026

    Recurring sync checks across the automation platform, the CRM and enrichment sources, plus a cleansing and deduplication standard for inbound lists. Failed syncs and duplicate spikes surface before a send, not after it.

    Marketo, Salesforce, webhooks, Python

  • 6sense intent activation

    2025–2026

    Buying-stage and intent signals turned into prioritized account segments that feed outbound and nurture, with segment definitions and refresh cadence written down, so sales and marketing argue about strategy instead of whose list is right.

    6sense, Salesforce, ZoomInfo

  • KPI and SLA reporting layer

    2026

    Process-level KPIs and SLAs for lead response time, data quality thresholds and routing accuracy, with each breach flagged to its owner during the week instead of at month end.

    Salesforce reporting, SQL, Sheets, API integration

Who I’ve worked with

48 of the companies I’ve delivered for through Get Levrg, across the US, UK, Canada and Australia.

One company, four roles

Get Levrg, Dhaka. From intern to revenue operations and CRM engineer in under two years, with project and client management alongside.

  1. Jul 2026 – presentGet Levrg, Dhaka

    Revenue Operation and CRM Engineer

    • Own data integrity and process automation across a live 12,000-record client system, rebuilding record-triggered workflows end to end, from first touch to closed transaction
    • Diagnosed a duplicate field silently zeroing scoring on 1,789 populated records, and corrected an archive query that had excluded 1,720 accounts from reporting. Neither was caught by standing review
    • Traced a silent data-loss path between client web forms and the downstream platform to a browser-side redirect, and eliminated it with a server-side webhook
    • Consolidated six competing automated processes into one before-save process with six branches, cutting per-save evaluations by 83%
    • Reduced active automation rules from 80 to 27 by removing conflicting triggers and redundant completion actions
    • Built multi-layer lead scoring engines that rank buyers by fit and intent, plus record-triggered flows that keep CRM data clean without manual upkeep
  2. Mar 2026 – Jul 2026Get Levrg, Dhaka

    Marketing Operations Specialist

    • Architected AI-powered research and validation systems that cut task time by 95% on datasets over 10,000 records, delivering validated, deduplicated data straight into client reporting
    • Built an AI validator that took per-record verification from 10 minutes to under 3 seconds at 98%+ reliability, replacing an entirely manual review
    • Scaled processing throughput 300% through AI-driven workflows while holding 99% data accuracy and cutting manual effort by 70%
  3. Mar 2025 – Mar 2026Get Levrg, Dhaka

    Business Development Associate

    • Ran discovery and requirement-gathering calls with clients across the US, UK, Canada and Australia, turning stated business problems into scoped technical work
    • Led delivery on CRM optimization engagements, including 12-month full-stack roadmaps for two enterprise accounts, both executed to plan
    • Led end-to-end execution for a global localization platform, scripting a Python tool that reached 98% accuracy and compressed a three-month manual process into six days
    • Used intent data and predictive analytics across LinkedIn and email outreach, holding 35%+ open rates and securing 15+ enterprise partnerships
    • Restructured North American funnel workflows to lift qualified meetings by 40%
  4. Dec 2024 – Mar 2025Get Levrg, Dhaka

    Business Development Intern

    • Sourced and validated 2,000+ records a month in Salesforce and LinkedIn Sales Navigator
    • Built automation tools that removed manual data entry, lifting tracking accuracy to 98% and speeding delivery by 30%
    • Began direct client contact with North American accounts
  5. 2025 – presentGet Levrg, alongside the roles above

    Interim Project Manager & Customer Success Manager

    • Managed 20+ client projects end to end: kickoff and onboarding, weekly execution calls, feedback sessions and final handover for US, UK, Canadian and Australian accounts
    • Guided internal delivery teams against weekly and monthly targets, and reported status directly to client stakeholders
    • Absorbed changing priorities and client concerns as they surfaced, keeping engagements on schedule
    • Designed and ran four internal culture and engagement programs alongside a full client workload

Paper and proof

A nineteenth-century botanical plate of a bitter orange branch: dark green leaves, white blossoms and the fruit
Citrus leaves, the subject of the paper. Köhler’s Medizinal-Pflanzen, 1887.

Published research, IEEE 2024

Optimizing Citrus Leaf Disease Detection: An Efficient Custom CNN Leveraging Training Parameters

A custom convolutional network that reached 97.84% validation accuracy and a 96% F1 score with far fewer training parameters than ResNet-50 or VGG16, built for resource-constrained environments. Presented at the 2024 IEEE International Conference on Artificial Intelligence in Engineering and Technology.

Read it on IEEE Xplore

Certifications

  • Klaviyo Digital Marketing Certificate
    Klaviyo    Oct 2026
  • Fundamentals of Predictive Project Management
    Project Management Institute    Sep 2026
  • HubSpot Reporting
    HubSpot Academy    Mar 2026
  • HubSpot Digital Marketing
    HubSpot Academy    Feb 2026
  • 6sense Digital Advertiser
    6sense    Apr 2025
  • 6sense Next-Gen Sales
    6sense    Apr 2025

Education

  • BSc Computer Science and Engineering
    BRAC University    2020 – 2024
  • Member, Robotics Club of BRAC University (ROBU)
    BRAC University    2020 – 2024
  • Higher Secondary Certificate, Science
    Adamjee Cantonment College    2017 – 2019

Languages

Bengali, native. English, professional working.

Platforms

  • SalesforceRecord-triggered architecture that holds its shape as the business changes underneath it, with rollups doing the arithmetic instead of a nightly script.Flows, Apex, SOQL, DLRS
  • MarketoNurture and lifecycle built so the next send does not start from scratch, and lifecycle stage means the same in the CRM as in the platform.Programs, smart campaigns, tokens
  • HubSpotPipeline design, property governance and workflow cleanup, reported in a way that survives a quarter-over-quarter comparison a year later.Pipelines, properties, workflows
  • Pardot and 6senseLayered scoring and intent activation, with segment definitions written down so the account list stops being a matter of opinion.Scoring, intent, segmentation
  • Python and SQLThe layer under everything else: validation pipelines, LLM-assisted checks, and the scripts that turn a three-month job into a six-day one.Validation, pipelines, LLM integration
Hatirjheel in Dhaka at night: light trails curving along the lakeside roads under a black sky
Hatirjheel, Dhaka. After dark here, a New York workday is starting.

Send me the symptom

Pick what’s going wrong and I’ll tell you what I’d look at first. Then write to me from here.

What’s going wrong?

What I’d look at first appears here.

I’d check whether two fields are writing the score. On one live system, a duplicate field was silently zeroing 1,789 populated records.

I’d read the report’s scope before the data. One archive query had quietly dropped 1,720 accounts from every report built on top of it.

I’d trace the handoff from the form to the CRM. A browser-side redirect was losing records; rebuilding it as a server-side webhook closed the gap.

I’d map every rule that fires on the object, in execution order. On one system, 80 rules became 27 and six competing processes became one.

I’d look for flows that rewrite contact roles. An overwrite guard keeps first-touch credit, and a scheduled retry catches records that arrive out of order.

I’d check whether an automation is re-sweeping historical records. That was the cause of one inflated counter I reconciled.

I’d write exit criteria for every stage first, then migrate live deals against an agreed old-to-new mapping.

Describe it in a few lines. I’ll reply with what I’d look at first.

Email me about this
siumibnakhter@gmail.com
Dhaka, Bangladesh. Working US, UK, Canada and Australia hours.LinkedIn and the IEEE paper