Manuel Vargas
ESVer en español
Menu

Engineer based in Colombia.Retail software for Latin American bodegas,GraphRAG for logistics teams,and language models trained from scratch.

Based in Colombia, working remotely in English and Spanish. Master's in AI and data science at Penn, in progress. Open to full-time roles and freelance contracts. I reply within 48 hours.

See the work
  • Bodega Boost landing page: 'The operating system for neighborhood commerce' over a dark montage of store owners

    Full-stack product. React and TypeScript

    Bodega Boost

    Bodega Boost
    Access on requestLive app
  • Iter AI sign-in screen with English and Spanish toggle and role-scoped tabs

    B2B platform. TanStack Start and React

    Iter AI Parking

    Iter AI Parking
  • Lante summary plate: entity graph motif with entity F1 0.933 against a 0.692 baseline, 93 annotated rows, 431 tests

    GraphRAG platform. Python and FastAPI

    Lante

    Lante
  • Training and validation cross-entropy loss for Sol's 52.9M-parameter baseline over 40k iterations, both settling near 1.3 nats

    ML training. PyTorch and TinyStories

    Sol

    Sol
  • Dawnveil summary plate: three parallel art pipeline lanes with the tools named at each step, peak monthly cost 82 dollars, 41 generated files, status planning

    Game, planning. Claude Code and Next.js

    Dawnveil

    Dawnveil

What are you hiring for?

Pick what you need on your team. I'll tell you whether I'm a match and point you to the work that backs it up.

Select one or more areas to see matching projects.

What I do

Product engineering

Multi-portal PWAs on Supabase with Edge Functions, billing, and persona-scoped RLS, built for LATAM unit economics. Shipped Bodega Boost (six-sided retail OS) and Iter AI Parking (four-portal parking platform).

AI and retrieval

GraphRAG copilots, hybrid Qdrant + graph retrieval, and tier-gated LLM features with cited answers and an evaluation harness. Lante (logistics Company Brain) and Iter AI Parking (Gemini owner copilot).

ML and data systems

Language models trained from scratch with honest ablations and leakage hygiene (Sol), plus large-scale anomaly pipelines with false-positive adjudication (Elephant in the Room, 509k contracts screened).

Let's work together

manuel@manuelvargas.dev

Open to full-time roles and freelance contracts. I reply within 48 hours.

Download resumeOr send a message through the contact form