EXPERIENCE, PERSPECTIVE & SELECTED WORK2016 — 2026
A LITTLE CONTEXT

Profile & perspective

01 / 05

My foundation is in analytics. My work increasingly connects the whole system.

I’m a data professional with a career spanning FIS, DecisionTree and EY since 2016. I turn business requirements into dependable reporting, connected pipelines and, more recently, structured AI workflows.

I like understanding what happens between the parts: how a business question becomes a metric, how an API response becomes useful data, and how an AI tool fits into a working application.

2016 — 2020FIS

Operations to analytics

2020 — 2021DecisionTree

Data & integrations

2021 — PRESENTEY GDS

Enterprise BI & applied AI

What the work shows

SELECTED EVIDENCE
03

Independent products. A broader perspective.

SocialMedia-Plus, Stockey and MyVisaFlow AI extend my experience into application workflows, system boundaries and AI-assisted development.

Explore the projects
PRODUCT WORK

I pay attention to the handoffs.

Some of the most interesting problems live where information changes shape. That is the thread connecting my work in reporting, integration and applied AI.

HOW I APPROACH MY WORK

Where I can contribute

BI and analytics engineering · Data integration · ICT business and systems analysis · Applied AI integration

gurpreet@gpsintown.com
PROJECT STUDY / 01
OPEN SOURCE · LOCAL AI WORKSPACE

SocialMedia-Plus

Make an AI-assisted social workflow repeatable, reviewable and connected to what actually happened.

CONTRIBUTIONIndependent product development
CORE STACKPython · SQLite · React · TypeScript · MCP
CURRENT SCOPELocal LinkedIn workflows

The problem

A desktop AI session can help write a post, but a useful publishing workflow needs more: durable context, editorial judgment, permission boundaries and a trustworthy record. A drafted action and a completed action are different states.

The system

A Python and SQLite ledger records requests, content versions, permissions and observed receipts. A React dashboard, CLI and local MCP interface expose that workflow to a desktop AI host. The host supplies reasoning and browser interaction; the workspace preserves state between sessions.

Desktop AI hostMCP / CLISQLite ledgerBrowser actionObserved receipt

Decisions that matter

  • Approval belongs to exact content. A fingerprint binds the content version, assets, destination, publishing authority and scheduled slot. A changed draft cannot silently reuse approval for a different version.
  • Uncertainty has its own state. Idempotency checks help prevent duplicate actions. A confirmed receipt needs observed evidence and a remote reference; uncertain actions enter reconciliation.
  • Review is independent by design. Five roles—practitioner, voice editor, busy reader, fair contrarian and relationship reviewer—receive the same frozen evidence packet. Supported hosts can delegate subagents; sequential fallback is explicitly identified.
Actual SocialMedia-Plus local dashboard
IMPLEMENTED SCOPE & VERIFICATION

The September 2026 release records 113 Python tests and TypeScript, build, dashboard, installation and MCP startup checks. These are project-reported results. LinkedIn is the current platform; other platforms and API publishing are planned. Local verification did not perform a live social publication.

Explore the open-source project

Public implementation, architecture and release notes available on GitHub.

PROJECT STUDY / 02
PROPRIETARY · IN DEVELOPMENT

Stockey

A strategy-engineering product built around the boundary between a proposal, a reviewed artifact and executable behaviour.

CONTRIBUTIONProduct direction & AI-assisted development
CORE STACKNext.js · React · TypeScript · Zod
ENGINEERING FOCUSShared contracts · exports · runtime safeguards

The problem

Strategy tools bring several distinct concerns together: describing an idea, reviewing its assumptions, generating an artifact and deciding whether it is eligible to run. My focus is making those boundaries explicit in both the product interface and shared contracts.

The architecture

The product combines a Next.js interface and documentation app with shared strategy-schema, broker-registry, AI-routing, generator and runtime packages. A Python export generator validates a strategy contract, sanitises archive paths and produces scaffold files with per-file integrity manifests.

Strategy definitionReview decisionsValidated scaffoldIntegrity manifestRuntime inspection

Engineering choices

  • Review decisions are explicit. Versioned proposals, per-finding decisions and stale-context checks keep an old review from silently applying to changed inputs.
  • Readiness is earned. The runtime foundation checks compatibility, package integrity, entitlement, lease state and matching dry-run evidence before reporting readiness.
  • Generated packages are inspectable. The generator uses file manifests and checksums, rather than treating an exported archive as an opaque result.

Working with specialist subagents

Documented development cycles split planning, bounded implementation and QA across roles covering UX, schemas, runtime contracts, security, broker research and release review. Review findings led to changes in stale-context handling, immutable decisions, diff provenance, lease and replay guards, redaction and mobile layouts.

The reviewed implementation milestone reports 76 tests across seven files, alongside documentation, type, build and three-viewport browser checks. These describe that milestone rather than a current independently rerun audit.

IMPLEMENTED FOUNDATION, WITH CLEAR LIMITS

The reviewed AI interface uses deterministic preview fixtures. The runtime foundation is a non-executing inspector, and generated live-order entrypoints are intentionally unimplemented. Live model calls, broker execution, production licensing and trading performance are not claimed.

Discuss the architecture

Independent proprietary work. Source repository is not shared here.

PROJECT STUDY / 03
PROPRIETARY · PRODUCT FOUNDATION IMPLEMENTED

MyVisaFlow AI

Connect guided intake, preparation checklists and report snapshots in a persistent applicant workspace.

CONTRIBUTIONIndependent product development
CORE STACKNext.js · PostgreSQL · Prisma · TypeScript
ENGINEERING FOCUSWorkflow state · scoped access · content provenance

The problem

Visa preparation is a sequence of interdependent decisions. An applicant’s answers affect the preparation pathway, required checklist, drafting context and next step. The product brings those pieces into a saved workspace, while separating applicant, agency and administrator responsibilities.

A connected preparation workflow

A catalog-backed intake matcher consumes saved answers and entry context, then returns a match, a request for more information or no match. Clear pathway identity is required before creating a workspace. Answer-conditioned checklists and drafting scaffolds support preparation, with progress persisted against the correct applicant and pathway.

Guided intakePathway contextSaved workspaceChecklist & draftsReport snapshot

Engineering beyond the screens

  • Versioned, explainable reports. A deterministic builder combines saved answers, checklist progress, template content, source-review status and paid entitlements into versioned database snapshots and printable HTML.
  • Access follows ownership and purpose. Applicant workspaces are owner-scoped. Agency lead views use scoped metadata and consent-dependent contact visibility; applicant answers and report details remain separated.
  • Payments reconcile on the server. Razorpay order and webhook flows verify signatures, reconcile payment events transactionally and update the correct pathway’s entitlement.
  • Content readiness is explicit. Source-review records, content versions and verification dates distinguish imported drafts from reviewed content. The rule-engine foundation gates production eligibility on published and reviewed state.

Evidence of delivery

The May 2026 project notes record 12 representative preparation scenarios, nine entitlement cases, report-snapshot checks, schema validation, type checks and a successful build. Earlier controlled regression checks cover owner and role boundaries, protected routes and checkout scope; they did not take real-money payments.

CURRENT IMPLEMENTATION & NEXT STEPS

The product foundation is implemented, with launch-readiness work continuing. Current matching and fulfillment are deterministic; model integrations are planned. Imported pathway content still needs human source verification. Document storage, report file generation and broader sharing remain future work. The product is presented here as software engineering work, not immigration advice.

Discuss the product

Independent proprietary work. Source repository is not shared here.

ENTERPRISE DELIVERY / EY
BUSINESS INTELLIGENCE · MIGRATION & VALIDATION

Preserve the meaning.
Modernise the reporting.

110+ legacy Qlik reports rebuilt and validated within a wider 700+ asset BI modernisation programme.

SETTINGEY Global Delivery Services
CONTRIBUTIONReport rebuilding, validation & UAT support
CORE TOOLSPower BI · SQL · DAX · data modelling

The delivery challenge

Moving reporting platforms is also a test of business meaning. Stakeholders need familiar metrics to remain dependable while reports, models and refresh processes change.

My contribution

  • Rebuilt and validated 110+ Qlik reports using Power BI and Microsoft-aligned data models.
  • Reconciled metric logic and reporting behaviour to support parity between legacy and migrated outputs.
  • Supported UAT, stakeholder sign-off and cutover readiness, connecting validation results to migration decisions.
  • Worked with metric definitions, data dictionaries, reconciliation checks and handover guidance as part of the broader delivery approach.

Why it matters

The work sits at the intersection of technical modelling and business trust. A visually complete report is only useful when its logic, data and acceptance criteria are understood by the people who rely on it.

SCOPE OF THE CLAIM

110+ reports describe my rebuilding and validation contribution. The 700+ figure describes the wider programme’s asset scope. Client data and confidential implementation details are not included.

ENTERPRISE DELIVERY / DECISIONTREE
DATA INTEGRATION · MARKETING ANALYTICS

Many sources.
A dependable picture.

Operationalising ingestion and reporting across 50+ source and connector types.

SETTINGDecisionTree Analytics & Services
CONTRIBUTIONSource mapping, ingestion, ETL & reporting
CORE PLATFORMSBigQuery · Redshift · MySQL · advertising APIs

The delivery challenge

Cross-channel marketing questions depend on data that arrives from different platforms, at different times and in different shapes. The integration work needs to connect source capabilities with business definitions and reporting expectations.

My contribution

  • Translated marketing requirements into warehouse schemas, source mappings, dashboard specifications and validation rules.
  • Built and operationalised ingestion and ETL across Google Ads, Google Analytics, Facebook and Bing Ads, FTP feeds, cloud files and warehouse tables.
  • Supported Google Ads Scripts and BigQuery pipelines, plus Tableau, DOMO, Power BI and Looker Studio reporting.
  • Documented API capabilities and dependencies, resolved production issues and supported post-release enhancements and technical handover.

The result

Integration and reporting automation reduced source-to-dashboard turnaround by an estimated 30–40%, as recorded in my résumé. Dashboards helped surface discrepancies, monitor campaign pacing and inform spend decisions.

SCOPE OF THE CLAIM

50+ refers to source and connector types. The turnaround improvement is an estimate, not an independently audited measurement. This case study describes delivery responsibilities without exposing customer data.