BHAVYA DABUR
Bhavya DaburBhavya DaburIIM Amritsar · IPM 01Open to internships, Apr–Jun 2027

I turn messy data into decisions. Finance, strategy and AI, built into tools people actually use.

NowStudying for CFA Level I

Now
IIM Amritsar, 5-year Integrated Programme in Management. BSc Quant Finance and Economics, CGPA 8.8
Won
S.P.A.R.C. National Case Competition, Hansraj College (DU)
Looking for
Internship, Apr to Jun 2027, Delhi NCR or remote
Scroll

Quant finance student at IIM Amritsar. On the Placement Committee I built our batch's recruiting machine from zero. On the side I run a one-person studio. I like problems with messy data, a real deadline, and someone waiting on the answer.

Selected work

01
      Problem

      What I built

      Result

      Placements Desk

      Problem
      44,000+ emails across shared inboxes (38,000+ sent, 6,000+ replies) and nobody could say where a single company stood.
      What I built
      A read-only pipeline that reads the committee's mailbox, works out the status and timeline of all 925 companies from the emails themselves, and publishes an encrypted, password-protected dashboard every morning at 08:00.
      Result
      193 live recruiter conversations in one view, and 45 dropped leads caught before they went cold. The committee runs on it.
      • Python
      • Gmail API
      • GitHub Actions
      • Cloudflare Pages

      Outreach Engine

      Problem
      A founding batch with zero recruiter relationships to inherit, and an internship season coming.
      What I built
      A 9,000+ lead pipeline (5,200+ BFSI contacts, 3,800+ HR heads) scored on fit, and a mailer that writes a personalised fit line for every company and threads the follow-ups. To decide whom to approach first, I mapped 566 companies from 21 other campuses' placement reports. When a third of the first wave bounced, I paused it and diagnosed all 730 failures by cause.
      Result
      3,000+ recruiters reached so far, with an hourly tracker for replies, bounces and follow-ups due.
      • Python
      • Apps Script
      • pandas
      • AI extraction

      The Strategy I Killed

      Problem
      A momentum backtest on the NIFTY 100 looked brilliant. Was any of it real?
      What I built
      A paper-trading system with a replay mode that cannot see the future, then five overfitting tests run on my own backtest.
      Result
      79% probability of overfit, and the edge died after brokerage and tax. I retired it and set go/no-go rules every future idea must clear. No proof, no trade.
      • Python
      • Backtesting
      • Deflated Sharpe
      • PBO

      Monte Carlo Risk Lab

      Problem
      A backtest shows one version of history. I want to know how bad things could get across thousands of possible futures.
      What I'm building
      A Python simulator that resamples real NIFTY daily returns into thousands of portfolio paths, then reads off Value-at-Risk, expected shortfall and the spread of drawdowns.
      Status
      In progress. The goal is a risk report I can stress-test, not a single number I have to trust.
      • Python
      • NumPy
      • pandas
      • Matplotlib
      Try the live simulator →

      AI Video Pipeline

      Problem
      A 34K-follower creator needed more short videos than one person could edit, and a community game members had stopped coming back to.
      What I built
      An automated reel pipeline: AI scene generation, local text-to-speech, captions and export. I studied 24 top posts to pick the format, audited the funnel into a 1,100-member community, and rebuilt its game as seven levels.
      Result
      Since I joined, the account has grown by 5,000+ followers and earned 500k+ likes, with reels produced end to end by the pipeline.
      • Python
      • Generative AI
      • Text-to-speech
      • Analytics

      What I bring

      1. Numbers I can defend.

        Econometrics, statistics, portfolio theory and option pricing from my degree; Excel and Python for the models. I'm most useful where a number has to survive scrutiny: valuation, risk, backtests and anything a senior person will question. I'd rather find the flaw in my own model before you do.

      2. A ranked list of what to do first.

        Breaking a vague question into a structure, sizing it with real data, and ending with a recommendation someone can act on Monday morning. Case competitions trained the framing; running outreach for a whole batch and selling to small businesses trained the follow-through.

      3. AI where it's strong, checks where it isn't.

        I use LLMs daily for code, messy text and first drafts, and I wire them into pipelines that run without me. The part I take most seriously is checking the output: AI is fast, not always right, and a team needs someone who knows the difference.

      I also run a one-person studio.

      I find local businesses cold, over calls and email, then build what they need and hand it over. Mostly websites and the WhatsApp and AI automation behind them, so an enquiry gets answered in seconds instead of days.

      It started with one real-estate website won on a cold call. Since then it has taught me what no classroom does: how to price, how to hear "no" and keep dialling, and how to deliver for someone who is paying with their own money.

      • Business websites
      • WhatsApp automation
      • AI automation systems
      • Cold sales
      • End-to-end delivery

      How I work

      01

      Write the brief first.

      Before I open Python or Excel, I write down the question, the inputs, the constraints and what done looks like. Most bad analysis starts with a fuzzy question.

      02

      Use the simplest tool that works.

      Plain Excel when it's enough, Python when the data gets big or messy, AI for code, unstructured text and first drafts. Never the fancy option just because it's there.

      03

      Try to prove myself wrong.

      Before I trust a result, I look for the reason it might be wrong: a leak in the data, a cost I ignored, a sample that's too small. I'd rather find the flaw myself than have a client or a manager find it.

      04

      Ship, then keep it running.

      The Placements Desk has rebuilt itself every morning since launch. A tool nobody has to babysit is worth more than a clever one that breaks.

      Python pandas SQL Excel Apps Script GitHub Actions LLM workflows Generative AI Automation
      Econometrics Regression Portfolio theory Option pricing Backtest validation Case strategy Cold sales Client delivery

      The road so far

      1. Class X, 91.4%

        Heritage Xperiential Learning School, Gurgaon.

      2. Class XII, 86.6%, and a fest for 2,000

        Led a five-person team that ran the school's annual fest.

      3. IIM Amritsar, founding IPM batch

        BSc (Hons.) Quantitative Finance and Economics. CGPA 8.8.

      4. Won a national case competition

        First place at the S.P.A.R.C. Society case competition, Hansraj College, University of Delhi.

      5. A studio, won one cold call at a time

        Started selling websites and automation to local businesses in Gurgaon.

      6. Placement Committee

        Elected after a multi-round process. Took on recruiter outreach and built the committee's tools.

      7. Your team, Apr to Jun

        Finance, strategy or AI. Delhi NCR or remote.

      Hiring an intern for April to June 2027 in finance, strategy or AI? Have a problem worth solving? I usually reply within a day.

      LET'S TALK