ai.agentroy.xyzAutomation & build — AI systems, backend, deploymentroy.mayaank@gmail.comagentroy.xyz ↗

Mayank Roy

AI Automation & Deployment

I automate the repetitive middle of back-office and operational work — rules first, models only where rules genuinely cannot reach, every decision auditable — and I ship it myself, front end to nginx. Twelve years in field operations and finance before this, which is why the automation matches what people actually do.

01

What I have built

Production systems running today, not demos. Several are live and clickable.

GL-coding pipeline

Python · SQL Server · pyodbc · rules engine · LLM fallback

Turns raw bank transaction exports into cleaned, coded, import-ready files for accounting systems. Deterministic SQL rules handle recurring transactions; a model call resolves only novel ones and writes its decision back as a new rule row.

The rule table extends itself — per-transaction cost falls as the system sees more data. Reconciliation logic blocks double-counting on liability accounts. Live demo →

Oilfield operations dashboard

Python · SQL · multi-source joins · web app

Pulls production, cost and operational records from separate systems into one executive view, with the joins and unit reconciliation that field data always needs before it can be compared.

One screen instead of four exports. Live demo →

Company Man dashboard

Python · vision extraction · local-first · nginx

A rig-site reporting tool for the company man. Photograph the tank board, the mud report, the morning numbers — the app reads them, returns a formatted report and a mobile dashboard the office can open. Runs on the company man's own laptop with a folder connection, or hosted.

Built from seven years of field experience with how rig-site data is captured and re-entered. Product page →

Transcription pipeline

Python package · yt-dlp · Whisper · CLI

Packaged tool that ingests long-form audio and video and returns structured, searchable transcripts — built to stop repeating the same manual extraction by hand.

A reusable package, not a one-off script.
02

Experience

I automate the repetitive middle of back-office and operational work — rules first, models only where rules genuinely cannot reach — and I ship it myself, front end to nginx.

2025 —
present

Data & Automation Engineer

Independent consultant · New Mexico
  • Built an end-to-end GL-coding pipeline that takes raw bank exports to coded, import-ready output — the manual step removed, not just sped up.
  • Hybrid architecture: deterministic rules handle recurring records, model calls resolve only novel cases, and each decision is written back as a new rule — model dependency falls as the system runs.
  • Scoped Azure OpenAI as the runtime target so client data stays inside firm compliance boundaries; constrained and validated every model output before it touches a ledger.
  • Ship and operate what I build: Ubuntu VPS, nginx reverse proxy, Let's Encrypt TLS, systemd services, git-push redeploys, live apps at cpa.agentroy.xyz and oil.agentroy.xyz.
2023 —
present

Founder & Lead Engineer

Web3UTXOS Technologies Pvt. Ltd. · DPIIT-recognized technology startup
  • Own the full stack: Python services, Linux server and cloud edge-server provisioning, Git workflow, code review and internal tooling.
  • Automated test suite as a release gate — 29/29 across four modules plus a 7/7 append-only log suite; nothing ships red.
  • Built internal Python tooling including a packaged media-transcription pipeline (yt-dlp + Whisper) to stop repeating manual extraction.
Feb 2023 —
Apr 2025

Power BI Report Developer / Data Analyst

EDGE Services
  • Designed and maintained dashboards over real-time operational data, with Power Query ETL and DAX models behind them.
  • Automated recurring reporting with Python and VBA, eliminating manual data-manipulation steps.
  • Trained non-technical field staff on the tools delivered — the skill that makes automation actually get adopted.
Aug 2017 —
Jan 2023

Wellsite Geologist / Mud Logger

Weatherford International · Paladin Geological Services · Selman & Associates · SE New Mexico & West Texas
  • A decade in field operations before software — which is why I can sit with the people whose process is being automated and understand what they actually do all day.
  • Owned data quality at the point of capture and reconciled real-time sensor streams against independent measurements.
Mar 2012 —
May 2014

Senior Systems Engineer

Infosys Ltd. · India
  • Built and maintained database-driven enterprise applications in SQL and PL/SQL; data extraction, transformation and reporting.
  • Delivered inside a large multi-team environment with formal ticketing, code review and release process.
03

Capabilities

Agent & LLM systems

Hybrid rules + LLM architecture · classification with rule write-back · prompt and output constraint design · tool/function calling · orchestrator–worker patterns with model tiering · Azure OpenAI and Claude

Python engineering

pandas · pyodbc · SQLAlchemy · Flask · packaged CLI tools · pytest suites as release gates · yt-dlp / Whisper pipelines

Front end

HTML, CSS and vanilla JavaScript · responsive and theme-aware layouts · dashboard and data UI · static and server-rendered pages · accessible, dependency-light builds

Deployment & ops

Ubuntu VPS provisioning · nginx reverse proxy · Let's Encrypt / certbot TLS · systemd services · git-push redeploys · Docker · DNS and subdomain management · SSH key management

Data layer

SQL Server, MySQL, PostgreSQL · schema design · ETL pipelines · validation and reconciliation · audit trails · append-only patterns

Integration

REST API design and consumption · JSON / CSV / XML · direct database connections · MCP and agent tooling · file and folder-based client workflows

Testing & quality

pytest suites run as release gates · integration tests against live databases · validation harnesses · reproducible environments · code review · documented standards

Domain translation

Twelve years in field operations and back-office finance — I can sit with the people whose process is being automated and understand what they actually do