0
mounting local volumes…
COPIED ✓

SUBJECT: ANURAG SINGH ROQUE · CLEARANCE: ON-PREMISE

ANURAG SINGH

//

@anuragroque · global username everywhere

I build AI that never phones home. I read systems the way I read people — quickly, completely, and a step ahead — then I build the machine that does it for you.

5 enterprise AI platforms, sole developer. 100+ users, 200K+ monthly transactions — every byte processed where it lives.

SEE THE WORK ↓ EMAIL ME
anurag@on-prem: ~/audit
$ whoami
anurag_singh // alias: roque · ai_engineer · privacy_absolutist
$ netstat --outbound --cloud
0 packets leaving local infrastructure # as designed
$ uptime --platforms
5 systems live · 200K+ txns/mo · 0 data leaks
SCROLL
0
Platforms shipped solo
0
Monthly transactions
0
Audit effort eliminated
0
Bytes sent to cloud

01 / PRODUCTS

Things I own.

Not client work. Not coursework. Products I conceived, built, and ship — each one runs entirely on your machine, because your data shouldn't pay rent on someone else's server.

★ PRODUCT · FLAGSHIP

FilloAI

The form filler that thinks. A Chrome extension that auto-fills any web form — structured fields and open-ended questions — using a locally running LLM. Zero cloud calls. It reads the page, reads your profile, and writes the answer before you've finished sighing at the question.

  • One-click resume upload: extracts your profile, drafts a cover letter, pre-generates 20 custom Q&As — all locally via Ollama
  • Intelligent field detection across name, id, placeholder, aria-label, and DOM context — dropdowns, radios, textareas
  • Works on LinkedIn, Indeed, Greenhouse, Lever, Workday, and any standard HTML form
  • Right-click any field to fill just that one, at the length you want
  • Everything lives in chrome.storage.local — no accounts, no tracking, no cloud
JavaScriptChrome MV3OllamaGemma3PDF Parsing
WEBSITE → DEMO VIDEO →
★ PRODUCT · DESKTOP

PulseDock Widget

A media controller that matches your mood. A sleek, frameless floating widget for Windows 11 that hooks directly into the Windows SMTC layer — real-time playback info and controls for Spotify, YouTube Music, VLC, Chrome, Edge, and anything else that makes noise.

  • Dynamic auto-theming — detects the active player and restyles itself to match its brand, plus custom dark/light/warm presets
  • 5 preset sizes from XS button-only to Large, with 20–100% opacity control
  • Async COM/MTA architecture: a dedicated QThread polls WinRT sessions without ever blocking the UI
  • SVG icons recolored at runtime; settings, position, and theme persisted locally
  • System tray controls, always-on-top, guaranteed clean shutdown — no zombie processes
PythonPyQt6winsdk / WinRTQSSPyInstaller
GITHUB →

02 / PROJECTS

Things I built.

Enterprise systems deployed in production, serving real users and real money. Each one designed, secured, and shipped by exactly one engineer. You're looking at him.

ENTERPRISE · RAGP-01

SOP Intelligence Portal

A privacy-first AI document assistant with multilingual support, deployed across 4+ enterprise environments serving 100+ users — fully on-premise, zero data leaving client infrastructure.

  • Hybrid retrieval: semantic embeddings + BM25 + cross-encoder reranking for both conceptual and exact-match precision
  • Real-time document updates with no model retraining
  • Multi-tenant RBAC, OTP auth, multi-layer prompt-injection filtering — hardened from day one
PythonFlaskPostgreSQLVector DBQwen/MistralBM25PWA
ENTERPRISE · AI AGENTP-02

Excellia AI

A spreadsheet validation platform powered by fully offline local LLMs, with an integrated chatbot for questions, analysis, and AI-assisted transformations — cutting analyst review time by 70% on 100K+ row datasets.

  • Threaded job queues, rule-based validation, and ML anomaly detection (Isolation Forest / Orange3)
  • Real-time spreadsheet preview and column-wise AI transformations
  • Flags inconsistencies and suspicious entries before a human ever has to look
PythonFlaskOllamascikit-learnOrange3Pandas
ENTERPRISE · FINANCEP-03

Limestone Reconciliation

Automated reconciliation for 10,000+ monthly transactions — eliminating 32+ hours of weekly manual finance work, an 80% reduction in processing time.

  • Memory-optimized pipeline handling 200K+ row datasets with psutil and memory-profiler
  • Real-time preview and automated discrepancy analysis
  • Self-serve transformation tools that freed analysts from engineering dependency
PythonPandasNumPypsutilML
OCR · AUTOMATIONP-04

KYC Data Automation

Multi-document OCR pipeline for Aadhaar, PAN, GST, and Passport with LLM + ML dual-mode name matching — cut Paytm's manual audit effort by 90%.

  • Google Cloud Vision + Tesseract OCR with SequenceMatcher bulk deduplication
  • Built to chew through large Excel datasets that used to take teams of humans
PythonTesseractGCV APIOpenCVSequenceMatcher
INFRA · IOTP-05

Self-Hosted Edge Server

Private cloud + NAS on a Raspberry Pi with Docker, Portainer, and Nextcloud, plus real-time IoT monitoring — temperature, motion, air quality, camera. Full Linux infrastructure on low-power hardware.

  • On-premise thinking applied past software, down to the metal
  • The same philosophy as everything else here: own the stack, own the data
Raspberry PiDockerPortainerNextcloudLinux
SLOT RESERVED
P-06

The next system is already in my head.
It just doesn't know it yet.

That's the public shelf. The full archive lives elsewhere →

VIEW MORE ON GITHUB → ▶ YOUTUBE — @ANURAGROQUE

03 / RECORD

Where I operate.

One title on paper. Five production systems in reality.

Data Analyst — AI & ML Engineering

TRPW Strategic Partners · Gurgaon, IN

MAR 2024 — PRESENT

Client Engagements

  • Paytm KYC Audit — OCR pipeline (Google Cloud Vision + Tesseract) with SequenceMatcher deduplication across large Excel datasets. Manual audit effort down 90%.
  • PRISM Workflow — automated Excel audit pipeline with Python, Pandas, NumPy, SQL, and Regex. Analyst cycle time down 60%.

Platforms Built & Deployed

  • Sole developer on 5 live enterprise platforms — Limestone, BetelTMS, BetelVMS, Excellia AI, SOP Intelligence Portal — serving 100+ users and 200K+ monthly transactions.
  • Local LLM deployments, RAG pipelines, ML anomaly detection, and OCR across all platforms — every system on-premise, zero data leaving client infrastructure.
  • Designed the multi-tenant RBAC, OTP auth, and prompt-injection filtering used across the entire fleet.

Education

Formal credentials, for those who collect them

Master of Computer Applications

Galgotias College of Engineering & Technology, Greater Noida
2021–2023 · CGPA 7.5

Bachelor of Computer Applications

Singhania University, Rajasthan
2018–2021 · CGPA 8.3

04 / ARSENAL

What I work with.

Tools are interchangeable. The pattern recognition behind them isn't. Still — here's the inventory, plus 300+ DSA problems solved for sport.

◆AI / ML

LangChainLangGraphRAGPyTorchscikit-learnOllamaHugging FaceLoRAspaCy (NER)EmbeddingsBM25Isolation ForestAI AgentsMLOps

◆Data Engineering

PandasNumPyPySparkSemantic SearchHybrid SearchCross-Encoder RerankingOpenCVTesseract OCRGCV APIMatplotlibSeabornOrange3

◆Backend

PythonFlaskFastAPIREST APIsGunicornNginxJinja2PWAJavaScriptC/C++

◆Databases

PostgreSQLMySQLMongoDBChromaDBRedisSQL

◆Cloud & DevOps

AWS S3 / EC2 / IAMDockerLinuxGitCI/CDSeleniumBeautifulSoup

◆Security

RBACOTP AuthSession ManagementPrompt-Injection FilteringInput Sanitization

05 / INITIATE CONTACT

You've read this far.
I already know why.

You have a system that needs building, data that can't leave the building, or a problem nobody on your team can crack. Convenient — that's exactly what I do.

+91 93195 31330 LINKEDIN GITHUB