// SIGNAL ONLINE — GOA ⇄ MUMBAI, IN

PRIYANSHU KUMAR

AI AUTOMATION ENGINEER · DATA QUALITY · FINANCIAL ANALYTICS

I build AI and data automation people can actually trust — the kind that runs unattended at 3 a.m. and doesn't wake anyone up. Currently teaching machines to double-check sovereign ratings data at MSCI. Dual-degree in Economics × Computer Science at BITS Pilani Goa, because one discipline felt like easy mode.

[ ENTER THE GRID ↓ ] [ OPEN A CHANNEL ]
SCROLL TO DESCEND

STAGE 02 // SYSTEM STATS

13
AUTOMATIONS IN PROD
all alive, none on fire
163
DATA CHECKS AUTOMATED
tiny robot inspectors, very strict
4,800+
ANALYST-HOURS SAVED / YR
≈ 2.3 full-time humans un-bored
4
PRODUCTION APPS SHIPPED
real users, real uptime

STAGE 03 // PLAYER PROFILE

Bilingual in balance sheets & build pipelines.

Dual-degree candidate in Economics and Computer Science at BITS Pilani, K.K. Birla Goa Campus — a polite way of saying I argue with both spreadsheets and compilers, and usually win.

My whole personality is trust in production: automation that survives messy inputs, LLM output that gets validated before anyone believes it, and audit trails clean enough to eat off. If it can't run unattended, it's not done.

Range: production data tooling → full-stack web platforms → the occasional computer-vision fever dream. I care about the craft — web development, machine learning, finance — and I am, admittedly, insufferable about data quality.

python automation LLM-assisted validation MCP servers financial analytics full-stack engineering
Portrait of Priyanshu Kumar
CITIZEN_ID PRIYANSHU KUMAR
CLASS: BUILDER · LVL 23

$ fortune
“The technology you use impresses no one. The experience you create with it is everything.”— Sean Gerety

STAGE 04 // QUEST LOG

Where I've been grinding XP

▸ MAIN QUEST — ACTIVE

AI Engineer · Practice School II

FEB 2026 — PRESENT

MSCI Inc. — Sustainability & Climate · S&C Government Ratings · Data Operations · Mumbai

  • Learned how sovereign & sub-sovereign ratings actually get made, end to end — then started automating the parts no human should still be doing by hand.
  • Shipped 13 production automations covering 163 screening factors. Analysts got roughly 4,800 hours a year back; I got a concerning taste for power.
  • Defensive web scraping, canonical country-name resolution, copy-on-write workbook handling, MCP server design, JSON audit reports — trust, but verify. Then log the verification.
  • Everything shipped at LOW-RISK under Data Operations AI governance (AIGC). My robots are certified chill.

▸ PREVIOUS QUEST — CLEARED

Full-Stack Development Intern

AUG 2025 — DEC 2025

Nirmaan Organisation — EdTech & Impact Hiring Platforms · Hyderabad

  • Delivered four production applications, including an Impact Hiring platform and an AI-integrated Resume Builder with role-based admin panels.
  • Hardened security with CAPTCHA, SQL-injection mitigation, and JWT auth — the holy trinity of "no, you can't do that."
  • Built on TypeScript, Next.js, Node.js, MongoDB, and the OpenAI API.

STAGE 05 // DEPLOY LOG

Shipped, live, and behaving

Four deployments on record. Open any panel for the case file.

DEPLOY_01MAR 2026

Nominal GDP Automation & Validation

An MCP server + validation pipeline that cross-checks nominal GDP figures for sovereign screening. The LLM gets tools with guardrails, the analyst gets JSON audit receipts, and silent mismatches get nothing.

PythonMCP ServerClaude API
OPEN CASE FILE →
DEPLOY_02FEB 2026

Sovereign Screens Web Scraper

A defensive scraper that survives everything that usually kills scrapers: layout drift, flaky responses, and countries that can't agree on their own name. Copy-on-write workbooks, because mutating source data in place is a crime.

PythonBeautifulSoupSelenium
OPEN CASE FILE →
DEPLOY_03MAY 2026

UN SDG Power BI Analytics Dashboard

Python ETL + an interactive Power BI dashboard over UN Sustainable Development Goal indicators — drill down by country, goal, and year. Built so a non-technical reader finds the story before their coffee cools.

PythonPower BIETL
OPEN CASE FILE →
DEPLOY_04APR 2026

Axiom Replica — Trading Terminal UI

A pixel-faithful clone of the Axiom trading terminal, streaming live market data over WebSockets. Dense, fast, real-time React — the UI equivalent of juggling knives, calmly.

ReactNext.jsTypeScriptWebSockets
OPEN CASE FILE →

RETRO ARCADE — earlier experiments in CV & games

Hand Gesture RecognitionMediaPipe · OpenCV
CV Volume ControllerOpenCV · MediaPipe
OpenCV Snake GameOpenCV · Python
Space ShooterPygame
Chat BotPython · NLP
HangmanPython

STAGE 06 // SKILL TREE

Points allocated across four branches

FINANCE & QUANT

speaks fluent balance sheet

Equity & index analysisPortfolio managementDerivativesRisk managementEconometricsTime-series analysis

DATA & ANALYTICS

pandas, no bamboo

PythonPandasNumPyscikit-learnSQLPower BIAdvanced ExcelMatplotlibSeaborn

AI & AUTOMATION

the robots report to me

Claude APIOpenAI APIMCP serversLLM-assisted validationAnomaly detectionETL automationWeb scraping

ENGINEERING

full-stack, no cap

ReactNext.jsTypeScriptNode.jsDjangoFlaskMongoDBMySQLPrismaGitDockerAzure

STAGE 07 // ORIGIN STORY

Character creation

BITS Pilani — K.K. Birla Goa Campus

M.Sc. (Hons.) Economics & B.E. Computer Science — Dual Degree

2021 — 2026

Chose the dual-degree difficulty setting: five years, two disciplines, one campus suspiciously close to a beach. Coursework across finance & economics, quantitative methods and econometrics, and core computing — plus a TensorFlow specialisation. The Econ half talks to analysts; the CS half ships to prod; I translate.

STAGE 08 // TRANSMISSIONS

Field notes from the grid

AI automation, data quality, and the occasional rabbit hole I fell into so you don't have to.

[ ALL TRANSMISSIONS → ]
AI · AUTOMATION
MAY 2026

Designing MCP servers analysts actually trust

Why audit trails, canonical resolution, and copy-on-write matter more than the model.

READ →
ROS · M1
JUN 2022

ROS on an M1 Mac with Multipass

A step-by-step guide to running the Robot Operating System on Apple Silicon.

READ →
THEORY
AUG 2022

Lambda calculus & lambda functions

A short, friendly tour of β-reduction and where Python's lambdas come from.

READ →

STAGE 09 // OPEN A CHANNEL

Let's build something trustworthy.

Recruiting? Collaborating? Want to argue about data quality? My inbox is an open port. Open to roles in AI automation, data quality, and quantitative finance. Response latency: better than my scrapers'.

kumar.priyanshu2k02@gmail.com