Imphal · Manipur · India

I'm Gary. Learning more of the job as I go.

I grew up in Imphal. I work in my family's business, the Punya Group: four dealerships (Hyundai, Kia, Toyota and Honda two-wheelers) and a logistics company we just started. I was appointed a director there. Until now I've been focused on the operational side: sitting with the work, then writing small tools so I can take the repetitive parts on myself. I'll keep picking up more as I go.

Punya Group of Companies BCompSc (Big Data) · University of Wollongong
Gary Mangsatabam Happy to chat
About

How I actually spend my time.

I finished a Bachelor of Computer Science at the University of Wollongong, majoring in Big Data, and came home to work in the family business. Until now I've been learning by sitting with people who already do the job, then taking that work on myself.

I started with payroll, monthly financials, parts inventory and OEM reporting. I run the tools I wrote. The point was to take that workload off the team, and I'll be expanding into more of the operation from here.

I like sitting with a messy process, finding the slow parts, and writing a small tool so nobody has to do it by hand again.

The companies

What's in the group.

Four dealerships, plus a logistics company we just started. Until now I've been involved across all of them, and I'll be taking on more.

Hyundai

Punya Motors

Hyundai dealership

Since 14 Nov 2025

Kia

Punya Automart

Kia dealership

Since 14 Nov 2025

Toyota

Punya Automotives

Toyota dealership

Since 14 Nov 2025

Honda two-wheelers

Punya Automobiles

Honda two-wheeler dealership

Since 14 Nov 2025

Punyagroup Logistics

Transportation, just started

Since 8 Jul 2026

Toolkit

What I use.

From a Computer Science degree, then from doing the work at the dealerships.

Languages
PythonJavaScriptTypeScript JavaC++R PL/SQLHTML5 / CSS
Web & data stores
FlaskNode.jsExpress MongoDBSQLREST APIs
Data & ML
Big Data (major)scikit-learnPandas NumPyPyTorchMatplotlib / Seaborn Model optimisation
Automation
SeleniumExcel pipelines (openpyxl) PDF generationTelegram bots PyInstallerGoogle APIs / OAuth
Business domain
Indian payroll (EPF / ESI / TDS)Dealer operations Financial reportingParts inventory OEM portal workflows
Working with people
TeamworkProject planning ResearchScrum / agile Git & GitHubAWS
Selected projects

What I've built.

Tools I use myself so the rest of the team doesn't have to sit through the same monthly grind, plus some university work. Company source stays private. Happy to talk through any of it.

Payroll platform illustration

Salary Automation Platform

Multi-brand payroll
Private

Monthly payroll for the four dealerships. I run it myself. It reads Excel inputs, applies each brand's rules, and writes slips, reports, bank remittance and compliance packs.

  • Per-brand rule engine for Hyundai, Kia, Toyota and Honda 2W in one codebase
  • Handles EPF, ESI, TDS, minimum-pay floors and workshop point pools
  • Preview and adjust screen so I can check numbers before the PDFs generate
  • Period / year-to-month reports built from saved month snapshots
Face animation deep learning illustration

Face Animation: A Deep Learning Approach

Final year project · UOW CSIT321
Capstone

Final-year team project. We built Edumate, a learning site where a talking avatar delivers lessons. You upload a portrait and the system animates it against recorded or synthesised speech. We spent a lot of time shrinking the models so they would actually run.

  • Compressed the mapping network 148.3 MB to 38.2 MB (−74%) and both generators by 50%
  • L1 unstructured pruning at 30% sparsity plus float32 to float16 casting
  • Optimised weights dropped in without changing the architecture
  • Full LMS around it: classrooms, assignments, quizzes, forums, video studio
  • Team of five, Scrum, four sprints
SVM confusion matrix showing 92.91 percent accuracy

Rice Variety Classifier

Support vector machine · university coursework
92.9% acc

A classifier that separates Cammeo and Osmancik rice grains from morphological measurements. University coursework, not a competition. Straight pipeline from a raw ARFF file through to a tuned, evaluated model.

  • 92.91% accuracy, 0.9292 precision, 0.9291 recall, 0.9292 F1
  • Grid search over kernel, C and gamma with 5-fold cross-validation
  • Standardised features and a train/test split
  • Confusion matrix above is the real output. 708 of 762 correct
Financial reporting illustration

Group Financial Reporting

Standardised monthly Excel format
Private

One monthly Excel layout for Hyundai, Kia, Toyota and Honda 2W, plus a group page so I can look at the four brands together instead of four workbooks that never matched. I fill this in myself each month.

  • One chart of accounts with four brand-specific overlays
  • Guided entry cells feeding print-ready income and position reports
  • Group insight page generated automatically from brand totals
  • Python builder regenerates every template deterministically
Parts inventory monitor illustration

Parts Inventory Monitor

Local dashboard from DMS exports
Private

A small local web app. I drop a read-only export from the parts system, and it gives me non-moving stock, a do-not-reorder list, and an audit sheet the workshop can print. It never writes back to the source system. I am the one who runs it.

  • Drop the export in a folder, run one file, open localhost. That is the whole workflow
  • Export-only, so nothing in the source system gets touched
  • One-click Excel file for the follow-up list
Portal monitoring bot illustration

OEM Portal Daily Monitor

Browser automation + Telegram alerts
Private

A daily script I run on my machine. It signs into a manufacturer's dealer portal, pulls the numbers I need, builds a PDF, and sends it to Telegram so I don't have to click through the site every morning. Optional Google Sheets and Drive sync. Packed as a Windows executable so I don't need Python installed on that PC.

  • Firefox automation with retries when the portal is slow
  • Telegram so the PDF lands on my phone
  • Optional Google OAuth for Sheets and Drive
  • PyInstaller build. No Python install needed on the machine
Hackathon puzzle solvers illustration

Hackathon Puzzle Set

Algorithms under time pressure
5 challenges

Timed challenges: Wordle, the Klotski sliding-block puzzle, a debugging round, and two build-from-spec problems. Useful practice at reading someone else's code quickly and making it pass.

  • Klotski solved with state-space search and visited-state pruning
  • Wordle clone with constraint filtering over the candidate word list
  • Debugging rounds: find and fix failures in code written by someone else
Experience & education

Where I've been.

Work

  • Punya Group of Companies

    Family business · Full-time

    Sep 2025 to present · Imphal, Manipur

    Until now I've been focused on payroll, monthly numbers, parts and OEM reporting, using tools I wrote so I can take that work on myself. I'll keep taking on more of the operation as I learn it.

  • AIESEC in Singapore

    Finance Department Associate · Full-time · On-site

    Sep 2023 to Jan 2024 · Singapore

    Five months in the finance function of a student-run global organisation. First real exposure to financial process and working to someone else's reporting deadlines.

Education

  • University of Wollongong

    Bachelor of Computer Science, major in Big Data

    Oct 2021 to Jun 2025

    Capstone: Face Animation: A Deep Learning Approach. Coursework across Java, Python, C++, R, PL/SQL, databases and machine learning.

  • Canadian International School

    IB Diploma

    2019 to 2021

    Arts Representative on Student Council. Involved in crowdfunding and volunteering initiatives alongside the diploma.

Representing the group

OEM dealer conferences.

I attended on behalf of the group. These are the meetings where manufacturers set targets, show new product and walk dealers through the year.

Simplified map of South and Southeast Asia with pins in Delhi, Bali and Singapore

Hyundai National Dealer Conference 2026

Bali, Indonesia · 12 – 15 March 2026

Honda 2W Dealer Conference

Delhi, India · 24 July 2026 · one day

Kia 2nd Half Dealer Conference 2026

Singapore · 5 – 8 August 2026

Contact

Let's talk.

If you work in dealer ops, payroll or just like talking about taking boring work off a spreadsheet, I'm around.

garymangsatabam@punyagroup.com