RaynnierGomez
EngineerSoftware ArchitectAI Builder
I build software systems, experiment with artificial intelligence, and study how agents and language models are changing the way complex software is designed, generated and maintained.
- Software
- the medium
- Engineering
- the mindset
- Artificial intelligence
- the current frontier
- Projects
- the evidence
02Trajectory
Five layers, one directionFour disciplines that look like career changes from the outside. From the inside they are one continuous expansion of the systems I am able to build.
- 01
Engineering
Formal engineering trained the reflex first: think in constraints, load paths and failure modes before thinking in features.
- 02
Software
More than ten years of development turned that reflex into working digital systems — backends, APIs, platforms, products.
- 03
Systems architecture
The interesting problem stopped being the code and became the shape of the whole: boundaries, contracts, data, failure and scale.
- 04
Artificial intelligence
Frontier and open-weight models arrived as a genuinely new material — probabilistic, general, and awkward to architect around.
- 05
Applied AI research
The current focus: agentic systems that can generate, modify and reason about real software, and the infrastructure needed to test that honestly.
I did not leave engineering for software, or software for AI. Each layer expanded the kinds of systems I could build.
03Selected work
[ 04 ]Projects are the argument. Each one exists to answer a question I could not resolve by reading about it.
- 01
Elean
Enterprise Low-code Environment with Autogenerative NucleusHow far can agents, structured domain knowledge and reusable skills go in generating real enterprise applications?
A system that combines domain models, agentic workflows and AI-assisted generation to produce containerized business applications. What exists today is a working generation pipeline; the long-term vision is stated separately from it, on purpose.
- agentic generation
- domain knowledge
- low-code
- containerized output
Active developmentCase study in preparation - 02
Local Agentic AI Lab
Agentic software researchCan agentic AI systems reliably generate, modify and reason about real-world software?
The research home for local language models, RAG, coding agents, context engineering, evaluation harnesses and multi-agent architectures. Experiments are published with their failures, not only their results.
- local LLMs
- coding agents
- RAG
- evaluation harnesses
ResearchCase study in preparation - 03
Player Development OS
Product and domain-modeling case studyWhat does a player's development actually look like as data, and who should own it?
A structured platform connecting players, coaches, academies, training, assessment, goals and evidence. The interesting part is not the sport — it is domain modeling, longitudinal data, analytics and consent design for athletes who are often minors.
- domain modeling
- backend architecture
- analytics
- privacy by design
In developmentCase study in preparation - 04
Personal AI Infrastructure
Persistent personal systemsWhat happens when an AI system becomes persistent infrastructure instead of a chatbot?
Ongoing experiments in autonomous agents, long-lived memory, retrieval, scheduling, tool use, voice interfaces and model routing — running against my own life as the test environment.
- autonomous agents
- memory
- model routing
- automation
ExperimentalCase study in preparation
04Research
Open threadsWork in progress, labeled honestly. A hypothesis is not a result, and an experiment that failed is still worth publishing.
- Hypothesis
Structure beats scale for generation
That well-modeled domain knowledge constrains code generation more effectively than simply reaching for a larger model. Testing this requires holding the model fixed and varying only the structure.
- Experiment
Evaluating coding agents on real repositories
Synthetic benchmarks reward the wrong behaviours. I am building harnesses that measure agents against real codebases, real test suites and real regressions.
- Open question
When does local inference stop being a constraint?
VRAM, quantization and context length define a hard envelope. The open question is which classes of agentic work fit inside it today, and which genuinely need frontier models.
- Research note
Context engineering is architecture
What an agent can see, when it can see it, and what it is allowed to forget are architectural decisions with the same weight as a database schema — not prompt tricks.
This is independent, self-directed applied research. No university affiliation, no doctorate, no institutional lab.
05Capabilities
Concepts, not logosGrouped by the kind of thinking involved rather than by tool. Specific technologies show up in the project case studies, where they mean something.
- Software systems
- Architecture, backend systems, web applications, APIs and distributed systems.
- Applied AI
- LLMs, agents, RAG, context engineering, inference, evaluation and model orchestration.
- AI infrastructure
- Local inference, GPU deployment, open-weight models, ASR/TTS, embeddings and experiment tooling.
- Product engineering
- Domain modeling, prototypes, SaaS architecture and product experimentation.
- Engineering
- Analytical modeling, structural thinking, quantitative reasoning and systems thinking.
06The Lab
Local inferenceReal hardware, real measurements.
A working bench for local inference: open-weight models, quantization, VRAM ceilings, throughput, ASR/TTS and the agent infrastructure that runs on top of it. The point is not that local models win — it is knowing exactly where they stop winning.
- Model
- Quantization
- Hardware
- Context
- Tokens/sec
- VRAM
- Method
- Date
Numbers arrive with the hardware, quantization, context length, methodology and date needed to interpret or reproduce them — or they do not arrive at all.
07Beyond code
The other discipline.
Photography is where I spend an equally large amount of time thinking about systems, observation and intention. Different medium, same habit: look closely, understand the mechanism, then decide what to keep in frame.
- Astrophotography
- Landscape
- Street
- Portrait
- Film
08Contact
One address. I read it.
Interesting technical problems, research collaboration, speaking and teaching. If you found this through a project, tell me which one.
No phone number is published on this site. That is deliberate — it lives only on the vCard behind the QR code on my business card.