I build AI, data, and automation systems that turn messy inputs into tools people can actually use.
- Building retrieval, validation, and scoring systems across AI and data workflows.
- Shipping repo-native tools that bring geospatial, finance, and ML work into the same decision flow.
- niche-radar ★0: A niche finder that pulls from external sources and turns noisy market inputs into signals you can use.
- ingo ★0: A shell-first RAG CLI for PDFs, OCR, embeddings, and vector search without a heavy app runtime.
- li-jobs ★0: A jobs collection API for scraping, normalization, and downstream job-data workflows.
- open_ksa ★3: A Python utility for working with Saudi Arabia's open data platform across thousands of datasets.
- tidyup ★0: A CLI utility for sorting files and folders by extension and date.
- translagent ★0: A translation app with transliteration and audio support, built end to end.
Private work: Purchase-Likelihood Scoring Pipeline · Geodatabase Validation Toolkit · ATS Scanner. More at works.estebanvalencia.com.
- Kimi K3 2.8T-A50B: the largest open model ever released; Opus 4.8-class at Sonnet 5 pricing (Latent.Space, 17 Jul 2026)
- Separating signal from noise in coding evaluations (OpenAI, 8 Jul 2026)
- Ollama: all aboard open models (Ollama, 9 Jul 2026)
Benchmark pulse: I check LiveBench daily.
Live artifact. Go + GitHub Actions rewrites this every 6h and scores 6 AI feeds for recency and relevance. Source · LiveBench