Matias Avendaño —builder, researcher^*,engineer.

Founder and ML/AI engineer from Lima. BS in Computer Science at UTEC, background in competitive programming. Raised $125k for teaching latinos how to speak English (didn't work out), but learned a lot.

Deep curiosity for algorithms and problem-solving. Experience on developing robust and scalable AI solutions.

Looking for high agency/technical engineering work and LLM research.

1. 01

   Pulso Salud · Naovi

   2025 — now

   Fullstack Lead

   Peru's largest occupational health network. 1M+ patients/year. Rebuilding the platform from scratch.

   • Modeled the whole occupational-health domain from scratch — ~120 Django models across 20+ apps (clinical, occupational, protocols, pricing, prediagnosis, forms, personnel, audit). Multi-tenant on django-tenants: Schema-per-tenant on Postgres, domain routing, RBAC, catalog seeding per business.
   • Re-architected the clinical protocol domain from a MariaDB schema where conditional logic lived as opaque LONGTEXT JSON blobs (7,340 protocols, ~191K rules, stringly-typed values with no referential integrity) into a fully relational Postgres model with boolean OR/AND condition trees, nested cases, typed exam modifiers, and proper junction tables, all backed by real foreign keys and exposed through a single atomic endpoint that builds the full protocol graph in one transaction.
   • AI protocol generator on top of that: Prompt chaining with structured outputs + pgvector embedding match (exam_matcher / embedding_service) to bind raw exam lists to risk factors and resolve conditional logic (gender, age range, exam type, scope level). Hours of clinical work to minutes.
   • QA agent for web platforms on AppRunner with the Claude Code SDK + Playwright MCP (now its better to use agent-browser).

   jan 20269 repos · 204 commits · 3d/barmay 2026

   Each bar represents this 3-day bucket.
2. 02

   Maxilar Labs

   2024 — 2025

   CTO · founding team

   AI agents running the administrative side of dental clinics: Reception, scheduling, finance.

   • Agents covering the full clinic back-office: Reception, scheduling, billing.
   • No-show rate 35% → 5% across 8 clinics, ~2k patients served.
   • Hard part: Orchestrating multiple agents over shared clinical state with WhatsApp as the only UI.
   • Won Startup Perú ($15k) Accelerated by UTEC Ventures IncUVa.

   oct 202410 repos · 379 commits · 3d/barjun 2025

   Each bar represents this 3-day bucket.
3. 03

   Amber

   2024

   AI Lead

   One of Peru biggest AWS partners. Set up the company's first AI team. Led 4 devs, set the playbook.

   • RAG over 25k pages with tables, images and diagrams. Real-time answers on WhatsApp.
   • Diff engine for 120-page government-bid docs, now used daily by the public-sector team.
   • Bid-scraping bot triaging hundreds of tenders/day into Slack (budget, personnel, licensing). Discovery time −30%.
   • Diabetes routine multiagent chatbot for 100+ users with doctor integration.
   • Set up the internal AI culture: Paper of the Month, Jailbreak Day.

   jun 202411 repos · 150 commits · 3d/bardec 2024

   Each bar represents this 3-day bucket.
4. 04

   Syntax

   2023 — 2024

   CTO → CEO · founding team

   Speech-to-speech English tutor for Latinos. Raised $125k, 5k installs.

   • Cross-platform speech-to-speech app. Helped people land jobs and gradsschool admits.
   • Built the AI memory layer (factual + everyday recall) that 3×'d retention. Pretty sure it was SOTA in its days. [talk]
   • Wore every hat: Frontend, devops, DS, mostly backend and AI.
   • Real challenge: Latency budget for real-time speech-to-speech without killing UX. Working with the models of the era (davinci, 3.5, 11labs in its early days). Everything was still being figured out. Cool times.

   feb 202320 repos · 706 commits · 3d/barapr 2024

   Each bar represents this 3-day bucket.