WELCOME TO MY STUDIO

Hey, I’mLoki.

I build websites and interactive tools.

Looking for a place to show your work, tell your story, or share what you’re building? I can help you put it together.

NOW, IMAGINE YOURS.

What would your website
look like?

Bring your projects, your story, or just an idea.
I’ll help you turn it into a website.

yourname.comLIVE PREVIEW
YOUR NAME®Selected work / About

A collection of things I’ve made.

Work worth
sharing.

Design. Research. Whatever you make.Let’s talk.

Portfolio — let your work do the introducing.

Your personal website for $100 USD. Domain & hosting: +$12 for an available standard .com; other domains are charged at actual cost. See what’s included

These are example directions. We’ll design around your story.

You’ll work directly with me, from our first conversation to launch.
Get to know me

STAY A LITTLE LONGER.

Want to try something I’ve built?

All labs

I build businesses.
And the systems behind them.

I work across business, analytics, and product development—from defining a problem and understanding the numbers to designing, building, and launching a solution. My work includes websites, interactive tools, dashboards, and AI workflows, with attention to both how they work and how people use them.

Currently building through Huvin Tech, including AuraMarc for professional personal branding.

NOW

Building through Huvin Tech

Developing AuraMarc for professional personal branding and working on client onboarding. Exploring website development, workflow automation, and analytics services for smaller businesses.

Details & skills
  • Product development
  • Workflow automation
  • Google Cloud (GCP)
  • Cloud Run
  • Vertex AI
  • Cloud Build
  • Artifact Registry
  • IAM & service accounts
  • Secret Manager
  • Docker
  • CI/CD
  • REST API integration
  • DNS & custom domains
  • HTTPS configuration
OCT 2025
— JUN 2026

ServCrust

Consultant · Mar–Jun 2026
Data Analyst Intern · Oct 2025–Feb 2026

Delivered an LLM-powered reporting platform used by 20+ senior leaders across three business functions, reducing reporting cycles from weeks to days.

Details & skills
  • Redesigned data requests into a self-service workflow, reducing turnaround from hours to minutes and removing approximately five weekly requests from engineering.
  • Built executive dashboards across three teams, bringing 8+ operational metrics into a shared reporting view.
  • Python
  • SQL
  • Amazon Redshift
  • LangChain
  • LangGraph
  • Langfuse
JUL 2024
— AUG 2025

Co-Founder · Gidorah Sports Management

Generated ₹10 lakh in revenue within the first two quarters and hosted 11 pickleball tournaments.

Details & skills
  • Used registration and retention analysis to revise pricing and tournament formats. Served approximately 100 paying customers and built a 190-member WhatsApp community through Paddle Out.
  • Business operations
  • Pricing
  • Community building
JAN–AUG
2024

Funding Consultant · Alien Technology Transfer

Screened 1,000+ med-tech startups for grant eligibility and funding fit, engaging 40+ C-suite executives and converting opportunities into five signed clients.

Details & skills
  • Defined client scope and objectives, then coordinated handovers to delivery consulting teams.
  • Odoo CRM
  • Client communication
  • B2B sales
  • Lead qualification
  • Stakeholder management
SEP 2022
— JUL 2023

Business Development Assistant · Paul, Weiss

Tracked court and board activity to identify 20+ qualified opportunities for partner outreach.

Details & skills
  • Prepared RFPs, pitch decks, and partner presentations that supported four new client wins.
  • Research
  • Microsoft Excel
  • Microsoft PowerPoint
  • Business development
01 / STATISTICAL MODELLING

European Football Match Forecaster

A Dixon–Coles forecasting project covering Europe’s five major leagues. Combines historical match data, automated updates, and walk-forward evaluation against baseline and bookmaker probabilities.

Data checks handle changing league structures, inconsistent team names, and invalid responses. The project documents where the model falls short of the market.

  • Python
  • pandas / SciPy
  • GitHub Actions
  • Backtesting
Explore the forecasts Read the code and evaluation
02 / PRODUCT DEVELOPMENT

AuraMarc

A product I created through Huvin Tech for professional personal branding. Current work focuses on developing the product and onboarding clients.

  • Product development
  • Personal branding
Visit Huvin Tech
03 / BUSINESS OPERATIONS

Gidorah & Paddle Out

Building a sports events business meant making decisions about pricing, tournament formats, registrations, and repeat participation.

A closer look at the business +

I co-founded Gidorah Sports Management and established Paddle Out Pickleball as its community arm. Across 11 tournaments, we served approximately 100 paying customers and built a 190-member WhatsApp community.

Registration and retention analysis helped identify where participants dropped off and informed changes to pricing and event formats. Revenue reached ₹10 lakh in the first two quarters.

A few things to play with.Go explore the labs
Data analysis · Product development · AI workflows · Cloud delivery
Analysis & data
  • Python
  • pandas
  • NumPy
  • SQL
  • Amazon Redshift
Reporting & applications
  • Power BI
  • Plotly
  • Microsoft Excel
  • Microsoft PowerPoint
  • Streamlit
  • FastAPI
AI workflows
  • LangChain
  • LangGraph
  • Langfuse
  • REST APIs
Modelling & delivery
  • Statistical modelling
  • walk-forward evaluation
  • GitHub Actions
  • automated testing
Cloud & deployment
  • Google Cloud (GCP)
  • Cloud Run
  • Vertex AI
  • Cloud Build
  • Artifact Registry
  • IAM & service accounts
  • Secret Manager
  • Docker
  • CI/CD
  • DNS & custom domains
  • HTTPS configuration
COMPLETED · 2022

Drexel University

BSBA · Management Information Systems and Sport Management

3.91 GPA · Summa Cum Laude
Dean’s List · Founder’s Scholarship

GRADUATE ADMISSIONS · CLASS OF 2027

UCLA Anderson

MS Business Analytics admit My choice

Carnegie Mellon University

MS Business Analytics admit

I chose UCLA, but visa appointment availability delayed my plans to begin.

Want a website of your own?

Explore personal websites

For business websites and analytics, explore Huvin Tech.

FOOTBALL / POKER / STRATEGY

Try my
projects.

Pick a project below. You can explore the model, try your own inputs, and read how I built it.

01Football Match ForecasterReal fixtures. Statistical forecasts. An honest benchmark.5 leagues · Python Dixon–Coles modelEXPLORE / CASE STUDY
Case study · Football Match ForecasterHow it was built
STATISTICAL LEARNING · FORECAST EVALUATION

Forecasting is only half the project.

A trained goals model should be judged against a strong benchmark, with its limitations visible alongside its predictions.

The question

How much of a closing-probability benchmark can a time-decayed goals model recover from public match history?

The original backend

Lokesh’s Python Dixon–Coles model fits attack, defence, home advantage and a low-score correction separately within each league, using penalised maximum likelihood. It is distinct from the player-based exploratory squad simulator.

The website integration

The original Python fitting code is retained. A versioned export supplies fitted parameters, cached fixtures, rating movement and the timestamped prediction log. The native frontend evaluates the original probability equations without embedding Streamlit.

Verification

Automated parity checks compare 60 forecasts and every cell of their 11×11 score distributions against the Python backend, including unseen-club fallbacks. Log summaries reconcile with the original Python scoring.

The evidence

The published walk-forward study covers 12,413 predictions. It improves on base rates but does not outperform normalised bookmaker closing probabilities in any league. The prospective log is displayed separately.

Limits

No injuries, lineups or cup congestion. Alternate fitted profiles are exploratory, not validated improvements. Feed, evaluation and rating-history dates are distinct, and exports must be refreshed and republished to update this website.

Original statistical backend and research: Lokesh’s Football-Major-5-Forecaster. Website integration and verification were built with AI assistance. Original code and evaluation

Loading the model snapshot…

02Football Tactics LabBuild a team. Explore the trade-offs.739 players · 2024–25 dataEXPLORE / CASE STUDY
Case study · Football Tactics LabHow it was built
SCENARIO MODELLING · DATA PRODUCTS

What changes when the team is yours?

Turn a static player dataset into a hands-on way to explore squad construction and modelling assumptions.

The question

Could a visitor build two real-player teams, change their tactics, and inspect both match outcomes and uncertainty?

The approach

Collected a 2024–25 snapshot from API-Football: 739 usable players from 74 clubs across five leagues. Preserved missing observations and selected one club/league spell per player to avoid blindly adding duplicate records.

The design decision

Separate observed player statistics from derived, group-relative indices. Use an explicitly exploratory engine for tactical effects, with linked shot, goal, block, and save events.

Verification

Automated checks reconcile team and player totals, validate unique lineups and goalkeeper placement, and verify repeatable simulations. Desktop and phone flows were exercised in the browser.

What this demonstrates

Data integration, model transparency, reproducibility, product interaction, and translating statistical outputs into something a visitor can investigate.

Limits

It is not a validated match predictor. Secondary-role fit is inferred from broad groups; the engine has no learned tactical interactions or cross-league adjustment.

Build process: AI-assisted implementation and testing, iterated from Lokesh’s product direction. No user-adoption or predictive-performance claims. Related project: original football forecaster

THE SQUAD BUILDER / 2024–25 EDITION

Real players.
Your alternate reality.

Two teams. Twenty-two decisions.
Build your XI, then find out what might happen.

2024–25 DATA

Loading the historical player pool…

Loading player data…

THE MATCH ENGINE

One match tells a story.
A thousand reveal the odds.

Custom squads · Neutral venue · No API calls per simulation

Choose players in both teams to begin.

Historical inputs. Simulated outcomes. This is an exploratory model of custom squads, not a validated predictor or a recreation of a real match. Player role flexibility is inferred from broad position groups.
03Poker Decision LabPlay the hand. Understand the decision.Equity · expected value · luckEXPLORE / CASE STUDY
Case study · Poker Decision LabHow it was built
PROBABILITY · DECISION QUALITY

Understanding a decision before the cards fall.

Make the difference between expected value and a single outcome visible and testable.

The question

How can a visitor understand hand equity, the price of a call, and luck without reading a page of equations first?

The approach

A standard 52-card deck, configurable known cards, and Monte Carlo sampling without replacement. An evaluator finds the best five-card hand from seven, including kickers and ace-low straights.

The design decision

Show outright wins and ties separately, while equity includes the actual fractional share of split pots. Recompute call economics instantly when chip amounts change, without rerunning the card sample.

Verification

Checked category counts for all 2,598,960 five-card combinations. Compared seven-card evaluation against all 21 five-card subsets for 1,000 sampled hands, and tested shared pots, duplicate prevention, and call arithmetic.

What this demonstrates

Algorithmic correctness, uncertainty communication, expected-value reasoning, and an interactive explanation of variance.

Limits

Unknown opponents are uniformly random, not betting ranges. Call analysis assumes one shared pot, no further betting, no rake, and no opponent folds. It is not a strategy solver.

Build process: AI-assisted implementation and testing, iterated from Lokesh’s product direction. Calculations run locally; no paid poker API is used. Related project: Poker Tracker

02 / THE POKER DECISION LAB

What are
your odds?

Choose your cards and run the odds.
Adjust the bet when you want to explore a decision.

Try an example hand +
THE WHAT-IF TABLE
THE FLOP
POT NOW 100
YOU
Your two hole cards

Click a card to change it. Card backs mean unknown.

NO API · CALCULATED IN YOUR BROWSER

Call analysis assumes no further betting, no rake, one shared pot, and no folding by opponents. Equity is conditional on the cards and opponent assumptions you set; it is not a complete poker strategy.

04Strategy LabA business problem. Millions of ways through.5 assignments · 9.7m configurations eachEXPLORE / CASE STUDY
Case study · Strategy LabHow it was built
COMMERCIAL JUDGMENT · CONNECTED FINANCIAL MODELLING

Make a recommendation. Show what must be true.

A fictional consulting case that connects diagnosis, strategic choices, operational constraints, and cash consequences.

The question

How do you turn an ambiguous business problem into a defensible recommendation, rather than a dashboard of disconnected metrics?

The approach

A fictional homeware company, three products, three channels, and five assignments. Visitors investigate the case, choose a plan and assumptions, and write four answers before submitting. Ten five-level decisions provide 9,765,625 selectable configurations per assignment.

The design decision

Connect price to demand, demand to capacity, sourcing to disruption and payment terms, and investment to operating costs, depreciation, working capital, and cash. Keep every assumption inspectable.

Verification

Checks cover financial reconciliation, five objectives per case, reproducible stress scenarios, score bounds, answer receipts, and duplicate-request caching. Live synthetic submissions tested reasoned writing, keyword lists, and instruction injection; these checks do not establish grading consistency for every possible response.

What this demonstrates

Problem structuring, commercial trade-offs, scenario analysis, resource allocation, sensitivity testing, and communicating an action plan.

Limits

Company data and response coefficients are fictional. Stress frequencies are illustrative; they are not real-world probabilities. AI feedback can vary. The educational score does not measure professional consulting ability. Tax, financing, and valuation beyond first-year cash are excluded.

Build process: AI-assisted implementation and testing, iterated from Lokesh’s product brief. No real client engagement, realised savings, or external validation is implied.

STRATEGY LAB / FICTIONAL CONSULTING CASES

Your recommendation.
Put to the test.

Meet the client. Work through the evidence.
Build a plan and defend your decisions.

Connecting to the case room…

Choose your assignment

How scoring and simulation work +

The outcome score is calculated by code. Each client objective earns up to six points, using average partial credit across 300 reproducible paired stress scenarios. Demand, costs, price sensitivity, execution friction, and named business stresses vary. Baseline and your plan face the same conditions.

Your assumptions affect a separate plan projection. The common stress set determines the 30 outcome points. Stress frequencies are illustrative, not empirically estimated probabilities.

Gemini assesses diagnosis (20), recommendation (25), assumptions and risks (15), and implementation (10). It must cite your writing. The server validates the response and adds the financial score; AI feedback may still vary between different submissions. Identical submissions can reuse a cached result.

A worked answer is one defensible approach, not the only correct plan. This is an educational score on fictional data. It does not predict professional consulting ability. No real client data, tax, financing, or lifetime valuation is included.

Choose your player.

GK stays in goal. Outfield players can cross position groups, with an illustrative role-fit adjustment. These are broad-group placements, not verified secondary positions.

CHOOSE A CARD

Make your move.

Cards already on the table are unavailable.