Portfolio · AI-native builder

Garima: eight things I designed, built and shipped with AI

I turn a business problem into a working AI product, often within a day. Each project below started from a real market question, runs on public data I sourced and checked, and was built end to end with Claude and Claude Code. All of them are live and free to use.

8live products, each built solo from a blank page
5.7Mofficial register records processed in one pipeline
0invented numbers: every fact is sourced, every assumption labelled
ClaudeAPI in the products, Claude Code to build them
Builder Multipliers: a community plan for a new kind of AI model
Community strategyWorking demo90-day plan

Builder Multipliers

Where a new kind of AI model, one that returns decisions with a confidence score instead of text, finds its next developers.

  • 10 sourced multipliers, from launch hackathons and SF meetups to MLH, TreeHacks and the n8n and Vercel communities
  • 6 build themes taken from real reported uses, plus a 90-day plan and the metrics to track
  • A working onboarding demo that routes new members to a track and hands low-confidence intros to a person

AI in the build: Claude to research and check every group and figure; Claude Code for the demo and the page.

HTML · CSS · JavaScript · public sources · Vercel View project ↗
Beyond the Labs: who buys expert data for AI after the frontier labs
Market sizingSQLBuyer map

Beyond the Labs

Who buys expert data for AI after the frontier labs: 12 vertical AI buyers in health, life sciences, law and finance, sized and sourced.

  • A SQLite database of accounts, investors, expert-evaluation programs and vendor revenue, with an open SQL console in the browser
  • Shows four vendors' reported figures (~$5.1B) exceed a $3.2B market estimate, and that vertical demand is small next to lab spend
  • A bottom-up sizing model with labelled assumptions, 5 channel partners and a 30-day test plan, built on 25 sources

AI in the build: Claude to source and check every funding round and figure; Claude Code for the SQL data model and the page.

SQL (SQLite / sql.js) · JavaScript · public sources · Vercel View project ↗
Signal Desk: monthly UK private-markets signal feed
Data pipelineAutomationGTM playbook

Signal Desk

A monthly feed of who is launching funds and setting up deals in the UK, pulled straight from the Companies House register.

  • Scans 5.7M register entries and finds 3,498 fund partnerships, carry vehicles and BidCo/TopCo deal stacks in 12 months
  • Groups them into 585 sponsor clusters, with an interactive monthly chart and a filterable feed
  • Refreshes itself every month through a scheduled GitHub Action; ships with a full outbound playbook

AI in the build: designed the classification rules with Claude, then used Claude Code to write and test the data pipeline and site.

Node.js · GitHub Actions · open data · Vercel View project ↗
Pipeline Truth: forecast health check for Salesforce exports
LLM productPrivacy by designPRD

Pipeline Truth

A free health check for sales forecasts: drop in a Salesforce export and see how much pipeline rests on stale data.

  • 5 rule-based checks and a value-weighted score, with stage, owner and forecast views plus a fix-list export
  • An optional Claude-written brief that only ever sees aggregates, with rep names replaced by labels
  • Ships with the full PRD: user stories, acceptance criteria, metrics and an AI decision log

AI in the product: Claude API behind a serverless function, with data minimised before it leaves the browser.

JavaScript · Anthropic SDK · Vercel functions View project ↗
Robot-Ready: go-live readiness kit for AI picking robots
Multi-tool appLLM memoRegulation

Robot-Ready

A go-live readiness kit for European warehouses deploying AI picking robots.

  • A 20-question readiness assessment and a 90-day go-live plan
  • A payback calculator built on Eurostat labour-cost data, plus a pilot KPI tracker
  • A plain-English guide to the EU Machinery Regulation that applies from January 2027

AI in the product: a Claude-written readiness memo that turns the user's own answers into a briefing for leadership.

React · Vite · Claude API · Vercel View project ↗
Pipeline from Scratch: outbound playbook for an AI-native CRM
GTM strategyScoring modelCalculator

Pipeline from Scratch

The first outbound playbook for a seed-stage AI-native CRM, written before day one.

  • 53 real target accounts from 2026 funding data, each scored and given a sourced "why now"
  • 4 message angles with 10 written messages, a 5-touch sequence and a channel mix
  • A working stack-cost calculator as a lead magnet, plus a market-pulse read for product

AI in the build: Claude for sourcing funding rounds and pricing pages, drafting the scoring model and copy; Claude Code for the site.

JavaScript · public funding data · Vercel View project ↗
UK Buy-to-Let Returns Lab
Financial modelOfficial dataWeb + Excel

UK Buy-to-Let Returns Lab

Personal name or limited company? A 10-year returns model for UK rental property.

  • Built on HM Land Registry prices, ONS rents and current HMRC tax rates
  • IRR in 13 currencies for overseas investors, plus a regional ranking
  • A matching Excel model, checked line by line against the web version

AI in the build: Claude Code to build and cross-check both models; every tax rule traced to GOV.UK.

JavaScript · Excel · open government data · Vercel View project ↗
Bid Multipliers: partner ecosystem for AI tender analysis in Sweden
Ecosystem strategyResearchReach model

Bid Multipliers

Who could bring an AI tender-analysis platform its next Swedish contractors, and why they'd say yes.

  • 14 partners in 5 groups, from tender-data platforms to trade associations covering ~12,600 firms
  • 3 integration pitches showing how data flows between a partner's tool and AI review
  • An interactive reach model and a 90-day partner pipeline, built on 23 public sources

AI in the build: Claude for multilingual (Swedish) research and source checking; Claude Code for the page.

HTML · CSS · JavaScript · Vercel View project ↗