Dhruv Nirmal.
Home
The path

How I got here.

A timeline of what I've built, newest first — production data and AI systems for enterprise clients, and my own products in between. Expand any step for the story, or open the full case study.

8 builds shipped Jul 2024 to now ~1 every 3 months 3 my own products 5 client systems
  1. Personal build May 2026 - now

    Facit

    A multi-tenant SaaS that unifies a café's POS, wages and 14 suppliers' invoices into one weekly profitability pulse.

    Facit turns the unintegrated data an independent café already generates — POS transactions, staff wages, supplier invoices — into a single weekly view of whether last week made money, why, and what to fix. It started as a real consulting engagement and was built as a white-labellable product rather than a one-off dashboard.

    Multi-tenant SaaSMulti-source integrationRevenue forecastingMenu engineeringBI
    View full project Turns unintegrated POS, payroll & supplier data into a Monday-morning decision
  2. Client work May 2026 - now

    Global spend-analytics platform

    End-to-end pipeline for the firm's largest account — ~6M invoice rows a week across 5 regions and 13 sub-regions.

    Own the end-to-end data pipeline for the company's largest client — spend data across 5 global regions and 13 sub-regions (~6M invoice rows plus PO, lookup and open-PO datasets; ~500–600k rows refreshed weekly).

    SQLPythonMicrosoft Graph APIAzureSQL Server
    View full project Weekly processing cut from ~5 hrs to ~75 minutes
  3. Personal build Jan 2026

    Job Hunt OS

    A single-user, file-based career-search OS that runs natively on Claude Code — an orchestrator plus specialist agents.

    A personal "Job Hunt OS" that runs natively on Claude Code with no paid API keys. It tailors ATS-friendly résumés to a job description and builds two-page PDFs, drafts cover letters, runs two-track networking, analyses résumé-to-JD skill gaps, and produces daily and weekly reports — all behind a human review gate.

    Multi-agent orchestrationClaude CodeMCPPythonPrompt engineering
    View full project Orchestrator + 3 specialist agents over a plain-file data model
  4. Client work Nov 2025

    Agentic report generation

    An agentic workflow that turns raw analytics outputs into structured client progress reports, rolled out company-wide.

    Built the agentic AI workflow that auto-generates structured client progress reports from analytics outputs and prepares them for sending. Rolled out company-wide and adopted by the entire analyst team, replacing a manual, per-client report-assembly process.

    ClaudeAgent designStructured outputsPython
    View full project ~12–15 hours saved across the team, every week
  5. Client work Hackathon winner Sep 2025

    LLM call-transcript analyzer

    A transcript analyzer that scores call quality, sentiment and expectations, orchestrated end-to-end from a drop folder.

    Built for an internal company AI-themed hackathon, where it placed 1st. A call/transcript analyzer that scores each engagement on several metrics — call quality, whether expectations were met, and tone — then routes the result to the right people. Later adopted into the company workflow.

    LLM workflowsAgent orchestrationSentiment analysisAutomationPython
    View full project 1st place, internal AI hackathon — since adopted
  6. Personal build Jul 2025

    Cloud data platform (IaC)

    Full cloud infrastructure provisioned as code — an automated ingestion pipeline from external APIs into Snowflake.

    A portfolio build that provisions a full cloud data platform as code and runs an automated ingestion pipeline into a Snowflake warehouse, with modular, version-controlled multi-environment deployment.

    AWSTerraformSnowflakeDockerPower BI
    View full project Reproducible dev / staging / prod from one codebase
  7. Client work Apr 2025

    ML spend-categorisation engine

    A one-vs-all text classifier that categorised ~A$12B of client spend across five years, with a Pareto split.

    Led the first time the firm ran this approach for a client: categorising a large spend dataset — roughly A$12B across five years — where only about 60–65% was reliably categorised to begin with. The result is a documented, repeatable pipeline now rolling out across the wider client base.

    scikit-learnTF-IDF / embeddingsLogistic RegressionFeature engineeringCross-validation
    View full project Cut a 1–1.5 month task to a single day's model run
  8. Client work Jul 2024

    Pricing & margin analytics

    Pricing and cost-leakage analysis across ~3,000 menu items for a multi-venue hospitality group, down to venue and supplier level.

    A large AU/NZ hospitality group could not see where price movement and cost leakage were eroding margin across thousands of line items. As sole analyst on the account I built the pricing analysis end to end and turned it into a decision tool for the category managers, presented to the Chief Procurement Officer.

    PythonSQLTableauPricing analyticsStakeholder reporting
    View full project ~30% (~A$2M) saved in a top category over a year
Let's talk

Want the detail behind any of these?

dhruvnirmal2111@gmail.com

Open to remote data & AI roles — Australia, India, and the Gulf / Singapore.