Jasper Nacho Martinez — AI Engineer and Oracle Developer Advocate. Seven years building ML systems, autonomous agents, and open-source tooling. I run the jasperan projects: Oracle AI Database forks, OSINT tools, and research on continual learning for LLMs.
I care about AI systems that keep working after the demo. That means memory, evaluation, and honest integration with storage — not another wrapper around a model.
Most of my time splits between (1) forking autonomous agents and swapping their memory layer for Oracle AI Database, (2) building OSINT and telemetry tools that scrape, log, and reason, and (3) teaching — workshops, keynotes, and DevRel content.
A portfolio of forks — oraclaw, picooraclaw, ironoraclaw, zerooraclaw, tinyoraclaw — each snapshotting a popular agent and swapping its SQLite/Redis/in-memory store for Oracle's Vector Search + Property Graphs.
Weighted ensemble voting, continual learning, and reasoning harnesses that don't collapse at 20 steps.
whatsapp-osint, waze-osint-tracker, and a handful of League-of-Legends ML tools. Practical, messy, shipped.
Conference talks, DeepLearning.AI course contributions, Oracle AI Developer Hub, and content pipelines.
Nine I'd actually point someone to — including two brand-new Oracle memory tools just shipped. Star counts are live from GitHub.
Logs WhatsApp online/offline transitions for any number in the world. The source of most of my "wait, you can do that?" reactions at conferences — and, quietly, my most-starred repo by an order of magnitude.
Curated technical resources for AI developers on Oracle AI Database and OCI. Notebooks, reference architectures, runnable examples.
End-to-end ML pipeline: Riot API extraction, feature engineering, real-time match predictions.
PicoClaw fork with Oracle AI Database as the sole persistence layer. Tiny, intentional, weirdly fast — flagship of the oraclaw family.
Expands standard LLM reasoning with structured planning and reflection loops. Benchmarks included, no hand-waving.
Load ONNX embedding models into Oracle AI Database with a single command — no Python runtime on the DB server, no glue code.
An AI memory layer for agents, backed entirely by Oracle Database. Inspired by memvid, rebuilt for a world where the DB is the vector store.
A stupidly easy RAG CLI for Oracle AI Database — Rich terminal UI, sensible defaults, zero YAML.
Real-time story drawing with StreamDiffusion, TTS, and ControlNet. Built for a live demo; kept for fun.
Sorted by GitHub stars, descending.
Logs WhatsApp online/offline transitions for any number in the world.
Curated technical resources for AI developers on Oracle AI Database and OCI — notebooks, reference architectures, runnable examples.
End-to-end ML pipeline: Riot API extraction, feature engineering, real-time match predictions.
PicoClaw fork with Oracle AI Database as the sole persistence layer — flagship of the oraclaw family.
Expands standard LLM reasoning with structured planning and reflection loops. Benchmarks included.
A stupidly easy RAG CLI for Oracle AI Database — Rich terminal UI, sensible defaults, zero YAML.
Real-time story drawing with StreamDiffusion, TTS, and ControlNet.
Load ONNX embedding models into Oracle AI Database with a single command.
An AI memory layer for agents, backed entirely by Oracle Database. Inspired by memvid.
An emotional state machine for agentic LLM pipelines — mood-weighted tool selection.
Oracle AI Database-powered document intelligence — SQL property graphs, vectorless RAG, knowledge-graph visualization.
Fast, small, autonomous AI assistant running exclusively on Oracle AI DB and Oracle Free Container.
Oracle AI Agent Harness — a fork of NousResearch/hermes-agent powered by OCI GenAI and Oracle 26ai Free.
JiT (Just-in-Time) and continual learning strategies for LLMs without catastrophic forgetting.
Near-real-time visualization of all world events logged on Waze — OSINT for traffic anomalies.
Tooling primitives for AI agents — loadouts, sandboxes, and structured tool interfaces.
Autonomous LLM agents that collaborate and vote using weighted ensemble decision-making.
WebRTC server that talks to Google Antigravity's chat — experimental client integration.
Code Abstract Syntax Tree (cAST) chunking — measurable gains over random chunk splits for code RAG.
League of Legends sniper-detection heuristic — check if someone's been hunting your account.
IronClaw fork with Oracle as the exclusive persistence layer — secure personal AI assistant.
Open-source transcription for audio files — Whisper-backed with a friendly CLI.
A fork of openclaw that uses Oracle AI Database for multi-agent assistance.
Multi-agent assistant + Oracle AI Database for persistent memory (TinyClaw fork).
Monitor Anthropic and OpenAI token usage on your local machine — no cloud, no leaks.
Showing 25 featured · 134 public repos total on GitHub.
All repositories on GitHubShort courses and workshops with learning platforms, usually on the agent-memory and Oracle AI DB side.
Co-instructor with Richmond Alake, Brendan Brown, and Andrew Ng. We show how to wire Oracle AI Database into agent pipelines for short-term working memory and long-term episodic recall. Notebooks, eval harnesses, and a playable end-to-end demo.
Nine international stages across three continents — mostly on agents, Oracle AI, and why your memory layer matters more than your model.
Talk · Production agent memory with Oracle AI Database
Talk · Vector Search, Property Graphs, and the Memory Stack
Keynote · Building the Brain and Backbone of Enterprise Agents
Workshop · Memory Engineering
Talk · Agent2Agent protocol integration in multi-agent RAG
Demo · Live experiential AI Foosball
Panel · The new operational complexity of agentic apps
Lightning · OSINT tooling lessons
Talk · Continual learning, honestly
Demo · Oracle AI DB + autonomous agents
Stages across 6 countries — Spain, Germany, France, UK, and the US (x4, including NYC in April).
Longform on Medium and Oracle DevPortal, video workshops, conference recordings, and the occasional podcast.
Short consulting engagements, workshop bookings, co-authorship on research, or just a weird agent idea you want to pressure-test. I reply to most things within a day.