mehanshlabsLet's connect

mehansh labs

Mehansh
Barthwal

Economist by training, builder by disposition. I write research on forecasting and policy, then go home and build the tools I wish I'd had while writing it.

Explore the labs
Open source
Research & ML
Builder

01 / about

An economist by training, an AI & tech obsessive by habit.

I spend my days reading econometrics and thinking about forecasting, and the rest of the time tinkering with language models, RAG pipelines, and small developer tools. The record below is the short version of how I got here.

State Bank of India

May 2026 to Present · 2 months

Analytics Intern

Active

State Bank of India

Building a hybrid econometric and ML framework to forecast SBI's monthly deposit and credit growth, benchmarking classical time-series models against regularised and sequence-based approaches, with macroeconomic and energy-price shocks layered in. Writing the production ETL pipeline that makes the whole thing repeatable.

TIES

Jul 2025 to Present · 1 year

Research Intern

Active

TIES

Contributing to the core research team on Indian political economy and public policy. Daily analytical briefs drawn from official data, legislation, and primary sources.

SVKM T&P Cell

Jun 2024 to May 2025 · 1 year

Student Coordinator

Completed

SVKM Training & Placement Cell

Primary liaison between recruiting companies and students. Coordinated campus placements and managed candidate interviews.

The Economic Transcript

Jun 2023 to Feb 2024 · 9 months

Editor

Completed

The Economic Transcript

Edited editorials, long-form magazine articles, and regular blog posts for a student economics publication. Maintained house style and managed the publication calendar.

SIDALCEAS EduTech

Aug 2023 to Sep 2023 · 2 months

Finance Intern

Completed

SIDALCEAS EduTech

Financial analysis and balance sheet work at an educational technology company. Contributed to team presentations on entrepreneurial ventures and key financial concepts.

02 / labs

Two benches, very different work.

Research & Academic Work

01 / 2

Formally produced in academic or research contexts — papers, internship outputs, and policy work.

AI Exposure and the Gender Wage Gap in India

Working Paper

Constructs an India-specific AI exposure index using sentence-transformer embeddings, merged with PLFS data for approximately 80,000 wage earners. Finds a robust 22% wage penalty for women, driven more by occupational sorting into automatable roles than by direct discrimination. Uses PSM, occupation fixed effects, and Oster sensitivity analysis.

  • Labour Economics
  • NLP
  • PLFS
  • Econometrics
  • India

Energy Shocks, Inflation Expectations, and Investment Frictions — A DSGE Model for India

Working Paper

Two small open-economy DSGE models extending Galí–Monacelli (2005) with India-specific features: separate crude oil and LPG shocks, RBI inflation expectations survey data, and Tobin's Q investment adjustment costs. Calibrated to quarterly Indian data, 2000–2025. Key findings include stagflation dynamics from oil shocks and TFP accounting for most output variance.

  • Macroeconomics
  • DSGE
  • Monetary Policy
  • Energy
  • India

Forecasting SBI Deposit & Credit Growth

Internship Project

SBI Analytics Internship · 2026

Hybrid econometric and machine-learning framework for monthly deposit and credit forecasting at the State Bank of India. Benchmarks classical time-series models against regularised regression and sequence-based approaches, with macro and energy-price shocks layered in.

  • Time-series
  • Econometrics
  • ML
  • Python

TIES Policy Briefs

Internship Project

Research Internship · TIES, New Delhi

Ongoing analytical briefs on Indian political, economic, and socio-economic developments for the Trade and Investment Exchange Society. Daily reports sourced from official data, legislation, and primary policy documents.

  • Public Policy
  • Research
  • Writing

Independent Projects

02 / 2

Things built out of personal curiosity.

02.5 / open for use

Built in the open. Ready to use.

A couple of tools I built and shipped publicly. If they solve a problem you have, feel free to use them.

Universal Scraping Architect

A Claude skill for building complete, production-ready scraping pipelines.

Who it's forUseful if you work with Claude and need to pull structured data from the web, PDFs, or APIs without writing brittle one-off scripts.

This skill lives inside a Claude-compatible environment and routes your scraping task across three modes — Firecrawl for dynamic sites, local Python for private data or static pages, and a hybrid pipeline for anything in between. Budget tracking, validation, and checkpointing come built in.

How to get it

  1. 1

    Head to the GitHub repo

  2. 2

    Navigate to engineering/universal-scraping-architect/ and open SKILL.md — that's the skill file Claude reads.

  3. 3

    Add the skill to your Claude environment — either by loading the SKILL.md into your Claude project context, or by cloning the repository and following the repo's skill-loading instructions.

  4. 4

    Tell Claude what you want to scrape. The skill will route the job, write the pipeline, and handle edge cases — you just point it at the data.

Open on GitHub

Shared AI Memory MCP

A self-hostable MCP server that gives Claude persistent, shared memory.

Who it's forUseful if you work across multiple Claude projects or accounts and want your AI to remember context between sessions — without it living in a single chat window.

AI clients forget things the moment a conversation ends. This MCP server gives Claude a durable external memory backed by your own Supabase database. You deploy it once, connect your Claude client, and from then on memory persists across projects, sessions, and even different accounts.

How to get it

  1. 1

    Clone the repo to your machine or server.

    git clone https://github.com/mehanshbarthwal-lab/shared-ai-memory-mcp
  2. 2

    Copy the environment template and fill in your Supabase credentials and a bearer token.

    cp .env.example .env
  3. 3

    Install dependencies and build.

    npm install && npm run build
  4. 4

    Run the server — the MCP endpoint is at /mcp and the health check is at /health. Deploy to Render, Railway, Fly.io, or any Node.js host.

    npm run dev
  5. 5

    Connect Claude (or any MCP-compatible client) to your endpoint using the bearer token. Memory is now persistent.

Open on GitHub

03 / research & publications

Research and working papers.

IJFMR · Vol. 7, Issue 2

2025

The Impact of Digital Payments on Retail Consumers and Small Vendors

A quantitative and survey-based study of how India's shift to digital payments has reshaped retail consumption patterns and small-vendor cash flow in urban markets. Combines transaction-level data with primary vendor interviews to trace the distributional effects of the payments transition.

  • Quantitative
  • Survey
  • Digital Payments
  • Policy
doi: 10.36948/ijfmr.2025.v07i02.41503View paper

Working Paper

Working Paper

2025

AI Exposure and the Gender Wage Gap in India: A Semantic Similarity Approach Using PLFS 2022–2024

This paper constructs an India-specific AI exposure index by matching official occupation descriptions against a curated corpus of AI-related texts using sentence-transformer embeddings, then merges this with worker-level PLFS data covering approximately 80,000 wage earners across two survey years. The central finding is a robust 22% wage penalty for women. While AI exposure is associated with higher wages overall, women concentrated in automatable, task-heavy occupations face larger penalties than those in more complex, human-complementary roles — suggesting AI may be widening existing gender inequalities primarily through occupational sorting rather than direct wage discrimination. Methodology includes propensity score matching, occupation fixed effects, a PSM-plus-FE hybrid, and Oster sensitivity analysis.

  • Labour Economics
  • AI Exposure
  • Gender Wage Gap
  • NLP
  • PLFS
  • India
  • Sentence Transformers
  • Econometrics

Working Paper

Working Paper

2025

Energy Shocks, Inflation Expectations, and Investment Frictions in a Small Open-Economy DSGE Model for India

This paper builds two small open-economy DSGE models for India extending the Galí and Monacelli (2005) framework with three India-specific features: separating crude oil and LPG as distinct energy shocks, incorporating household inflation expectations from the RBI's Inflation Expectations Survey of Households, and modelling investment adjustment costs via Tobin's Q. Both models are calibrated to quarterly Indian data from 2000 to 2025. Model 1 uses crude oil as a single cost-push shock; Model 2 uses a composite energy index weighted 75% crude oil and 25% LPG, reflecting how differently the two fuels transmit through the Indian economy. Key findings include stagflation dynamics from oil shocks, TFP shocks explaining the majority of output variance, and UIP risk-premium shocks driving most of the exchange rate variance — with direct implications for monetary policy responses to energy-driven inflation in open economies.

  • Macroeconomics
  • DSGE
  • India
  • Energy Shocks
  • Monetary Policy
  • Inflation Expectations
  • Small Open Economy
  • Calibration

04 / kit

Tools I use.

Technical

  • Python
  • R
  • Stata
  • SQL
  • Git

Data & ML

  • pandas
  • scikit-learn
  • Time-series
  • Forecasting
  • RAG
  • Generative AI

Research & Analytical

  • Econometrics
  • Economic forecasting
  • Panel data
  • Policy research
  • Desk research
  • LaTeX

Tools & Workflow

  • ETL pipelines
  • PDF extraction
  • Jupyter
  • AWS basics
  • Editing
  • Graphic design