I'm an MBA candidate in Business Analytics at BITS Pilani, with a bachelor's degree in Physics from Banaras Hindu University. Most recently, I was a Product Management Intern at a health-tech company.
Scroll down to see my experience, the projects I've built and why I chose an MBA. Have a question? Ari, my AI assistant, can walk you through any project or tell you why I'd be a good hire.
Projects across risk, analytics, strategy and Gen AI products
Top 20
of 5,000+ teams, FinSpark '26 banking cybersecurity hackathon (Jul 2026)
1st Runner-up
“The Hot Seat”, Interface '26, BITS Pilani (Sep 2026)
PM Intern
i3digital health: AI voice agent for physicians, forecasting model and competitor dashboard
About
Why a physics graduate builds data products
You're probably wondering why a physics graduate is building dashboards and data products. Here is the short version of a calculated, and genuinely enjoyable, journey.
I fell in love with physics while preparing for engineering entrance exams, so much that, after clearing a few colleges, I dropped engineering altogether and pursued a B.Sc. in Physics instead. Which raises the obvious question: why an MBA, and not an M.Sc. or a PhD?
What drew me to physics is how it lets you imagine: how a moving electron generates a magnetic field, how matter shows a dual nature. Imagination has no bounds, but physics adds rules and constraints that make an idea hold up in the real world. Imagine freely, then test against reality. That is the habit I wanted to keep.
With a foundation in first principles, I wanted to point that thinking at something other than machines. Could it create real impact in business? That question brought me to an MBA in Business Analytics at BITS Pilani.
The best way to find out was to start doing it. The projects below are the result.
Product & AI
Discovery and specialist interviews, TAM/SAM/SOM sizing, PRDs and use cases, and Gen AI products built with RAG, LLM agents and guardrails that keep numbers grounded.
Risk analytics
Credit scorecards (WoE/IV), default prediction and risk tiering, cost-based cut-offs, and explainable fraud detection with XGBoost and SHAP.
Markets & volatility
GARCH and GJR-GARCH, VaR and Expected Shortfall, Monte Carlo simulation, portfolio optimisation and forecast evaluation against simple benchmarks.
Experience & education
Where I've worked and studied
May 2026 – Jul 2026
Product Management Intern
i3digital health · Oncology AI
AI Voice Agent for Physicians: led discovery through 10+ specialist interviews, sized the opportunity with TAM/SAM/SOM and turned the findings into a 5-use-case PRD. Built it with the engineering team across timeline, UI and database design; the agent answers clinical queries via RAG, decides when and where to fetch data before responding, and redacts patient PII.
Forecasting Model: built an expert-weighted Python model that benchmarks PD-1/PD-L1 drugs on sparse data (10–20 entries per biomarker) and forecasts competitor movement, which informed go-to-market strategy.
Competitor Analysis Dashboard: built an ETL pipeline and automated dashboard that surfaces companies' drug-trial funding slowdowns and the current treatment landscape for each biomarker, highlighting white-space opportunities for leadership.
2025 – 2027
MBA, Business Analytics
BITS Pilani
Coursework in quantitative methods, analytics, marketing and corporate finance. Coordinator of the Creative Club.
2020 – 2023
B.Sc. Physics
Banaras Hindu University
Mathematical modelling, statistics and first-principles problem solving.
Leadership & achievements
First Runner-up, “The Hot Seat” — Interface '26, BITS Pilani · Sep 2026
Recovery strategy for a quick-commerce dark-store network (SwiftCart), backed by a network model, presented as a 6-slide pitch and defended in a live CEO Q&A.
Top 20 of 5,000+ teams — FinSpark '26 · Jul 2026
National banking cybersecurity hackathon by Bank of Maharashtra with the Indian Banks' Association and DFS, Govt. of India. Cross-silo fraud detection engine with XGBoost, SHAP explanations and 4 independent agents with plain-English LLM reasoning; selected for the Grand Finale at COEP Pune.
Coordinator, Creative Club — BITS Pilani · Nov 2025–present
Led a 5-member creative team producing posters, social media creatives and event branding for 10+ department events. Conceptualised and led “Say It Without Saying It”, a national-level, product-based marketing competition with 300+ participants.
1st Runner-Up — FinQuest, IIM Rohtak · Nov 2025
Finance quiz competition, among 200+ participants.
Projects
Selected work
Each project starts from a business question and uses real data unless noted. All figures come from the project outputs.
AI product · RAG on Cloudflare
Ari — portfolio RAG assistant
The assistant in the corner of this page. It answers recruiters' questions about my work using retrieval over a curated knowledge base, so its answers rest on real project material rather than a generic persona. It runs on Cloudflare Workers, Vectorize and KV with multi-provider LLM fallback, and has a live analytics dashboard for questions, latency and visitors.
Live analyticsquestions, latency and visitor tracking
Internship · i3digital health
AI Voice Agent for Physicians
Led discovery through 10+ specialist interviews, sized the opportunity with TAM/SAM/SOM and wrote a 5-use-case PRD. Built with the engineering team as an agent that answers clinical queries via RAG, decides when and where to fetch data before responding, and redacts patient PII.
10+specialist interviews
5PRD use cases
DiscoveryPRDRAGPII redaction
Gen AI product & architecture · Jul–Oct 2026
CollectIQ — AI Receivables Briefing for Indian SMEs
Designed and prototyped a B2B2C AI product that turns an SME's invoice ledger into a plain-language weekly brief on who owes what and whom to chase first, with no accounting-system migration. An agentic collections layer escalates overdue invoices and drafts reminders for the owner to approve. Code computes every figure and the LLM only writes the text; go-to-market runs through CA firms.
4-tierescalation ladder
B2B2Cvia CA firms
Pythonscikit-learnStreamlitAPScheduler
Prototype
Gen AI & NLP · Agentic RAG · Jul–Oct 2026
Earnings Intelligence Copilot
Agentic RAG over earnings calls: a retrieval-augmented pipeline over 30 earnings-call transcripts across 3 sectors that extracts and scores management guidance on capex and margin. Chunking, embedding and dense retrieval in Python, with a tool-calling agent layer in an agent-to-data pattern for governed enterprise question answering.
30transcripts
3sectors
PythonRAGEmbeddingsLLM agents
Market sizing & segmentation · Apr–May 2026
ServZone — Managed Marketplace for Urban Car Servicing
Bottom-up sizing of an urban car-servicing market, stress-tested against service-volume and penetration scenarios, with customer segmentation and a positioning thesis that prioritises the highest-value service categories. Demand is forecast with a value-chain ratio method and extended to a 3-year view.
₹60 CrYear-1 opportunity
3–4%share, ~1.25 lakh visits
~₹105 CrGMV by Year 3
ExcelMarket sizingSegmentation
Gen AI product · Agentic RAG
FeaturePilot — Copilot over Product Feature Requests
A copilot that lets a PM ask what users want most, what a competitor offers and what to build next, across public GitHub issues for Metabase and Apache Superset. Counting tools produce the numbers, retrieval cites real issues, and a validator rejects any figure no tool returned. It is built on the tested ReviewPilot engine.
6agent tools (ReviewPilot engine)
PythonSQLiteStreamlitLLM agents
In progress
AI agents · Orchestration
Prashant AI OS
A personal multi-agent system for recurring tasks. The first agent, ARES, retrieves structured information on request over Telegram. Further agents and a performance dashboard are planned, prioritised around hardware limits.
A competitive-intelligence dashboard for pharma strategy teams that maps where each competitor's drug trials stand: completed, active and new initiations, and which cancer targets they cover.
A verified professional network for oncologists. Trust is tied to medical licence IDs, a graph-based engine suggests peers by specialisation, and direct messaging supports referrals and consultations.
Compares a WoE scorecard, a random forest and XGBoost on 30,000 card customers, prioritising recall of defaulters. It then builds a cost-based cut-off, 10 calibrated risk bands, SHAP explanations and PSI stability checks.
0.552PR-AUC (XGBoost)
50.2%of defaulters in top 2 bands
SQLXGBoostSHAP
Fraud detection · Hackathon
FinSpark — Cross-Silo Fraud Detection Engine
Built for Bank of Maharashtra's FinSpark '26 hackathon. A correlation model across transaction, account and channel silos, using XGBoost with SHAP explanations, to catch fraud that looks clean inside any single silo. Trained on simulated data.
A joint project on a simulated 260,000-loan book across four product lines. It includes a PostgreSQL ETL, a PD scorecard with WoE/IV selection, five risk tiers, and vintage, DPD and early-warning KPIs exported to Power BI.
4product lines
5risk tiers
PostgreSQLPythonPower BI
Simulated data
FX risk · Monte Carlo
Currency Risk Quantification — GJR-GARCH
USD/INR risk on a $1M exposure using 10 years of FRED data. The model is chosen by testing for asymmetry (GJR-GARCH or GARCH), then 10,000 Monte Carlo paths give 95% and 99% VaR and Expected Shortfall.
Tests what can and can't be forecast. ARIMA return forecasts tie the historical mean, while GARCH volatility forecasts beat a 30-day rolling estimate: returns are hard to predict, risk is not.
p = 0.98ARIMA vs mean
3.6% betterGARCH, p = 0.005
PythonARIMAGARCH
Portfolio construction · Excel
Portfolio Risk Dashboard
Can an optimised portfolio of Indian assets beat the Nifty 50 on a risk-adjusted basis? It computes minimum-variance and max-Sharpe weights and a 20-point efficient frontier, delivered as an Excel workbook where every KPI is a live formula.
20frontier points
ExcelPythonscipy
Product strategy · Market sizing
QuickReach — Quick-Commerce Expansion into Tier-2 India
Should a quick-commerce player expand into tier-2 India, which cities first, and can a dark store break even? A bottom-up Excel model of 10 cities built on Census, HCES and TRAI data, with Low / Base / High scenarios, four customer personas and a positioning thesis built on top-up categories.
₹980 CrYear-1 market (base)
~705orders/day to break even
ExcelPythonopenpyxl
Customer analytics · Retention
Customer Segmentation, Survival & Retention
Segments 7,043 subscription customers with K-means, tests churn drivers and models time to churn with Kaplan-Meier curves. One price-sensitive segment holds two-thirds of revenue at risk. Delivered as a 14-sheet Excel dashboard and a Tableau view.
$1.56M/yrrevenue at risk
47%churn in top-risk segment
SQLscikit-learnlifelinesTableau
Operations · SLA risk
IT Service Operations & SLA Risk
A real ServiceNow log of 24,918 incidents. Handoffs drive SLA misses: the miss rate rises from 21% with no reassignment to 82% at four or more. An at-open early-warning score and four sized operational fixes follow.
44%of misses in top 20% flagged
AUC 0.77at ticket open
PythonXGBoostSurvival analysis
Project analytics · Monte Carlo
Project Delay Risk Analysis
Measures schedule and cost overruns across 115 completed real projects, tests an early-warning signal and sizes schedule buffers with a 10,000-run Monte Carlo on the activity network. Recommends reference-class contingency.
40%finished on time
p = 0.005early-warning signal
PythonEVMMonte CarloPCA
Skills
Tools and methods
Product
Product discovery & user interviews
Market sizing (TAM/SAM/SOM)
PRDs, use cases, product design
Competitive analysis, go-to-market
PowerPoint storytelling
AI
RAG pipelines and vector search
LLM agents and tool calling
Guardrails: code computes, LLM narrates
Prompt engineering
Cloudflare Workers, Streamlit
Data
Python, SQL, R
Power BI, Excel, Tableau
EDA and data visualisation
ETL pipelines and dashboards
Statistical data analysis
Machine learning
XGBoost, random forest, logistic regression
SHAP explainability
Credit scorecards (WoE / IV)
GARCH volatility, Monte Carlo
Survival analysis, K-means, PCA
Let's talk
I'm an MBA candidate (class of 2027) open to full-time roles in risk analytics, data science and product. The quickest way to reach me is email. You can also ask Ari, the assistant in the corner, about any project on this page.