Juan Camilo Martinez
Applied AI Engineer
I build audio/speech ML pipelines and multi-agent systems. Each project below explains what I built and why.
Get in touchAudio/Speech ML
Production pipelines with Whisper, wav2vec2, Parselmouth, and Silero VAD. Pronunciation scoring, prosody analysis, phoneme alignment.
Multi-Agent Systems
Agent orchestration with LangGraph, Redis event bus, guardrails, health preflight, and observability dashboards.
Business-Aware AI
Unit economics at the architecture stage. Pricing models, evaluation systems, and cost-per-inference thinking from day one.
Featured Work
The two systems that define my engineering identity.
Invoz
Audio ML pipeline built from 46 research papers. Scores spoken English across 11 dimensions (7 acoustic + 4 linguistic) using Whisper, wav2vec2, Parselmouth, Silero VAD, and LLM-based coaching.
Holus Observatory
Live observability dashboard for a 32-agent autonomous content system. Watch agents reason, generate, review, and publish content across platforms in real time.
Social Media Automatization
Multi-platform publishing API and content pipeline. One thought in, platform-native posts out — X, Threads, LinkedIn, Instagram, Facebook. Powered by the Holus agent system.
More Projects
Production systems across AI, data, and full-stack.
Pilaster
Version control and memory system for AI generation workflows. Tracks iterations with intent notes, parameter diffs, and failure pattern warnings across ComfyUI and multi-backend pipelines.
AI Advisory Board
Multi-agent reasoning system where specialized agents debate strategy, sales, and market trends to reach collective decisions. 3rd place, Datadog Hackathon.
Holusight
Hybrid document retrieval system combining BM25, vector search, and reciprocal rank fusion reranking with Claude-powered answer synthesis.
Job Tracker CRM
End-to-end job search CRM with Kanban workflow, AI-generated cover letters, resume tailoring, and pipeline management. Used personally to manage an active job search.
Genpeli
Local-first AI video editing pipeline. Smart cuts, word-by-word captions, audio normalization, and social media delivery for short-form content.
Personal Projects (7)
How I Built This Chatbot
The engineering behind the portfolio chatbot — LLM-as-judge evaluation, RLHF feedback loop, live context injection from WHOOP/Strava, and autonomous knowledge expansion.
Fitness Dashboard
Real-time health analytics integrating WHOOP and Strava APIs with natural language querying via RAG-style contextual retrieval.
Focus Time
macOS menu bar timer for hyperfocus awareness. Periodic check-ins during deep work sessions with focus pattern review.
Astoria Conquest
Geospatial routing and map visualization to run every street in Astoria, Queens.
Accountability Partner
Public commitment tracker for daily morning workouts. Every session visible to everyone.
NLP Data Science
Interactive presentation on pronunciation error detection and NLP analysis using Quarto and Python.
HRV Research
Machine learning analysis using WHOOP data to identify key drivers of daily HRV and optimize recovery protocols.
