Apurwa Sarwajit
Apurwa Sarwajit

Apurwa Sarwajit

Technical product manager. I build 0 to 1 AI products for enterprises and stay close to the code, working in Python, SQL, and TypeScript and publishing open-source packages on npm, including an MCP server. Most recently I led AI Studio at Redblock.ai, where security teams build and deploy browser agents for the legacy, non-API systems that APIs never reached, with eval harnesses gating every release.

I have spent most of my career in new categories, where the hard part is finding the right problem, shaping the product, and proving it works before it scales. AI Studio crossed $1M in contracted revenue in year one. Earlier I launched India's first pre-onboarding Fraud Score, adopted by the top 6 banks and reaching $4M ARR in 24 months, on a decisioning platform running 50M+ transactions a month.

Outside work I ship side projects end-to-end with Claude Code and Codex, building, breaking, and fixing small things to understand how AI is changing the way software gets made.

Work

Redblock.ai Product Lead, AI Studio 2025 - Present BureauID Senior Product Manager, Fraud Score 2023 - 2025 BureauID Product Manager, Platform 2022 - 2023 View full experience

Selected products

AI Studio Browser agents for enterprise IT and security $1M Fraud Score AI risk score for banks and fintechs $4M ARR Score Engine Configurable decisioning platform 50M+ tx/mo View all products

Education

IIT Roorkee B.Tech, Electrical Engineering 2013 - 2017

Skills

  • Browser agents
  • Agent orchestration
  • Eval harnesses
  • Enterprise GTM
  • Pricing
  • Python
  • SQL

Product work

AI Studio

Browser agents for enterprise IT and security

Took an on-prem agent platform from idea to production so enterprise teams could build browser agents for legacy systems without stable APIs.

Users
IT and security teams
Traction
4 banks and enterprises, $1M contracted
Built
redflow, pipelines, deployments, eval harnesses

Fraud Score

Fraud Score (Mule Score) for banks and fintechs

Launched a pre-onboarding fraud score that turned proprietary models and graph signals into a revenue-generating risk product.

Users
Banks, fintechs, lending teams
Traction
$4M ARR, India's top 6 banks
Impact
170K+ fraud accounts, $12M losses prevented

Score Engine

Configurable decisioning platform

Built the configurable scoring layer behind BureauID's fraud and verification products across clients, channels, and risk use cases.

Users
Risk, fraud, and onboarding teams
Scale
35+ clients, 50M+ monthly transactions
Impact
Bulk Verification drove 60% of revenue

RTO Score

Return-risk scoring for D2C commerce

Built a commerce risk score to help brands decide which orders were likely to return before shipping.

Users
D2C commerce operators
Scale
70+ brands, $10M/month GMV protected
Impact
Reduced return-to-origin rates 15%

1-Click Checkout

B2B2C checkout for Shopify and WooCommerce

Launched a checkout product for merchants that reduced buyer friction across Shopify and WooCommerce storefronts.

Users
Shopify and WooCommerce merchants
Motion
B2B2C checkout product
Impact
Reduced checkout drop-off 25%

Experience

  1. Feb 2025 - Present · Bangalore

    Founding Product Lead, AI Studio

    Redblock.ai

    • Lead a 12-person product and engineering org, owning AI Studio strategy, roadmap, delivery, and design-partner feedback loops.
    • Took AI Studio from 0 to 1, shipping production browser agents across 4 banks and enterprises for use cases including application onboarding, credential rotation, and ITSM ticket execution.
    • Landed a Tier-1 US bank and a top global card network as anchor design partners; shaped initial use cases and converted POCs into $1M in contracted revenue in year one.
    • Defined core AI Studio primitives, AI Skills, Agents, Pipelines, and Deployments, so teams could compose tools, knowledge, actions, and control flow into reusable pipelines for IGA, PAM, and ITSM workflows.
    • Created redflow, a patent-filed proprietary DSL that lets users pass intent and skill-specific instructions to agents with execution guardrails; enabled new AI Skills to be created and deployed in under 60 minutes.
    • Productized browser / GUI agents that execute tasks across enterprise web apps without stable APIs, with eval harnesses for workflow accuracy, safe tool use, and regression testing before rollout.

  2. Nov 2023 - Feb 2025 · Bangalore

    Senior Product Manager, Fraud Score

    BureauID

    • Launched India's first pre-onboarding AI fraud score, generating $4M ARR in 24 months with adoption by India's top 6 banks and leading fintechs.
    • Identified 170K+ fraudulent accounts, reduced false alerts 71%, and prevented $12M in fraud losses using proprietary models and graph features.
    • Improved 6- and 9-month retention 40% by refining ICP, use cases, pricing, and sales channels through 12 months of PMF experiments.
    • Increased product margins 47% in 15 months by replacing third-party data vendors with internal data products and proprietary risk signals.
    • Expanded into Southeast Asia and new-to-credit lending, adding bank and mid-market lender segments while supporting 4M monthly transactions.

  3. Dec 2022 - Nov 2023 · Bangalore

    Product Manager, Platform

    BureauID

    • Built BureauID's proprietary Score Engine, a configurable decisioning platform powering fraud and verification products for 35+ clients and 50M+ monthly transactions.
    • Launched Bulk Verification and onboarding tools that drove 60% of revenue, cut turnaround time 70%, improved time-to-value 15%, and reduced onboarding TAT 25%.

  4. Oct 2021 - Dec 2022 · Bangalore

    PM, RTO Score & 1-Click Checkout

    Blaze by BureauID

    • Built RTO Score, a risk-scoring product for D2C commerce that reduced return-to-origin rates 15% and protected $10M/month in GMV across 70+ brands.
    • Launched 1-Click Checkout, a B2B2C checkout product for Shopify and WooCommerce merchants that reduced checkout drop-off 25%.

Building

Open-source and side projects I ship end-to-end with AI coding agents, mostly around agents, MCP, and developer tooling.

TypeScript · MCP

gdocs-mcp

Open-source MCP server for Google Docs & Sheets. Self-hosted with local OAuth, so tokens never leave your machine or touch a third party.

GitHub npm

Node.js · CLI

cc · Command Center

A CLI that orchestrates parallel Claude Code agent sessions across projects, driving each in its own tmux pane so several agents work at once.

View on GitHub

TypeScript · Agents

browser-agent-chat

A web chat app for driving a live browser agent, with a streamed view of the browser so you can watch the agent work in real time.

View on GitHub

Security · Agents

SafeAgents

An assurance layer for AI agents: adversarial agents pentest an enterprise agent before production and issue an independent security report.

Building now

React Native · Edge AI

Split

A local-first expense tracker for couples with a fully on-device LLM agent, running a quantized Qwen model on the phone itself. Partners sync over the home intranet, end-to-end encrypted. No cloud, no accounts.

Building now

C++ · Hardware

esp32-home-monitor

Reproducible ESP32-S3 board bring-up and a BME280 home-environment monitor. A hands-on hardware side quest, documented end to end.

View on GitHub

Education

Indian Institute of Technology, Roorkee

B.Tech, Electrical Engineering

Elected to three consecutive student-government roles, culminating as President of the Students' Affairs Council. Represented 10,000+ students to the administration and drove governance reforms, including a landmark 24/7 campus-access initiative.

2013 - 2017

Skills

AI / Agents

  • Browser agents
  • Agent orchestration
  • Tool use
  • Eval harnesses
  • Agent reliability
  • Agent observability
  • Open-source LLMs

Product

  • 0 to 1 enterprise products
  • Design partner programs
  • Enterprise GTM
  • Pricing & packaging
  • A/B testing
  • Discovery & research
  • Roadmap & OKRs

Technical

  • Python
  • SQL
  • Reads production code
  • REST APIs
  • Prototypes agent evals
  • Ships internal tools
  • Claude Code & Codex

Thinking about

How browser agents become reliable enough for enterprise workflows.

How AI products should be evaluated after launch, when real users, edge cases, and messy workflows show up.

How risk products turn messy behavioral signals into decisions teams can trust.

How to build products that feel simple while hiding operational complexity.

Say hello

I like meeting people who are thinking about AI, products, enterprise software, risk, agents, or just interesting problems. If something here connects, reach out. I am always up for a good conversation, a sharp idea, or a small thing worth building together.