#2
Sanlo

AI Hackathon Journey

A five-month sprint series at Sanlo, where I built four experimental MVPs and three production prototypes across HR, Fintech, and VC—automating resume screening, natural-language transfers, KPI extraction, and more in record time.

Astro Next.js Supabase LangChain Vercel AI SDK shadcn/ui Python CrewAI Firecrawl Crawl4AI Stripe API Supabase Auth

Intro

When Sanlo’s CEO challenged us to prove AI’s business value, we answered with a sprint-driven hackathon series. Over four one-week experimental MVPs, we tested ideas—from resume summarization to natural-language transfers—then shifted into three one-month production MVPs in the VC space. Spanning roughly two months, this journey taught me how to move from zero to prototype at lightning speed and then scale those prototypes into robust, user-facing tools.

What I Did

  1. Experimental Sprints (4 MVPs)

    • Week 1 & 2: Built HR (CV summary, fit-scoring, interview Qs) and Fintech (NL bank transfers with Stripe & Supabase Auth) prototypes in under five days each.
    • Week 3 & 4: Extended to two more MVPs, experimenting with Astro → Next.js RSC, Python, and CrewAI, and overcoming anti-bot web-scraping challenges using Firecrawl and Crawl4AI.

  2. Extended VC Projects (3 MVPs, 1 month each)

    • KPI Extractor: Parsed pitch-deck PDFs and scraped websites—navigating anti-bot defenses—to extract key metrics and auto-generate executive summaries.
    • Document-Driven Report Generator: Let users upload arbitrary docs; AI-powered prompts then answered guided questions and composed a coherent, tailored report.
    • Spreadsheet Structurer: Transformed unstructured gaming-industry spreadsheets into normalized tables, giving investors clear performance dashboards.

Throughout, I owned the frontend—leveraging Astro and Next.js, real-time Supabase sync, and shadcn/ui components—while wiring in AI via the Vercel AI SDK, LangChain pipelines, and Python where heavy lifting was needed.

Key Takeaways

  • Sprint-Driven Agility: Four experimental MVPs in four weeks validated ideas rapidly, even without measuring user metrics.
  • Product-Grade Refinement: Three month-long sprints produced medium-to-high impact tools—users gained real value, though most chose to stick with legacy processes rather than pay for new tools.
  • Technical Versatility: Python’s AI ecosystem proved powerful for heavy processing; JavaScript/React stacks excelled in rapid UI iteration.
  • Prompt Craftsmanship: I learned firsthand how small prompt tweaks can drastically improve AI outputs.
  • Mindset Shift: Shipping small, learning fast, and focusing on solving the core problem beats waiting for perfection—every prototype was a step toward clarity and impact.