Piyush Gupta

Engineering teams lose hours to context-switching between review, testing, and merge cycles. I built a multi-agent SDLC platform (BUILDER / TESTER / REVIEWER) with hash-chained audit trails and blast-radius gates that block high-risk changes automatically. The platform manages its own development — every PR ships through the same pipeline it improves.

5× faster PR cycle · ~$3–6/ticket · blast-radius gates

Comparing hotel prices across Booking, Airbnb, Expedia, and 3 other platforms meant 6 tabs and 20 minutes of manual work per trip. I built a local-first meta-search engine that scrapes, normalises, and ranks results from all 6 platforms in a single query. Cross-platform price intelligence cuts pre-trip research from 20 minutes to under 2.

6 platforms, 1 query

Manual job-hunting — parsing JDs, tailoring resumes, and tracking applications — consumed hours per week. I built an end-to-end pipeline that sources roles, runs an LLM fit-check against my profile, and generates a tailored resume and cover letter per role. The system cut active application time by ~80% while increasing targeting precision.

~80% reduction in active application time

Most personal finance tools require manual categorisation or expensive bank integrations. I built a local pipeline that ingests raw bank-statement CSVs, auto-categorises transactions with an LLM, and produces a monthly narrative summary alongside spend charts. Zero manual entry — all insight, no data-entry overhead.

Zero manual entry

Health data lives in silos — Apple Health exports, sleep-tracker CSVs, and workout logs never talk to each other. I built an aggregation agent that unifies four data sources and surfaces a weekly LLM digest: trends, anomalies, and one actionable recommendation. Single weekly read replaces four disconnected dashboards.

4 sources → 1 weekly digest