Research, images,
software and automation.

Hand me reports, manuals, websites or exported records along with the question. I compare the sources, keep supporting passages with the findings and distinguish what is known from what is inferred or still unanswered.

Sources can disagree without either being useless. I look at their dates, scope and authority, carry those distinctions into the answer and leave unresolved questions visible. A software manual for the wrong release or an inferred email address needs to be identified as such.

• Research briefs with citations and open questions
• Company and contact research that separates published facts from guesses
• Technical document analysis tied to a specific software version

I can take a person, product or idea from reference photos through generation, selection and targeted edits. I give each reference a role, such as identity, pose or visual treatment, then judge each result against the references and the rest of the set.

The detailed work is in the comparisons: a face that has drifted, a pose that no longer fits, a prop that changes between scenes. I use saved references, targeted edits and transparent pieces to develop the series without starting every revision from scratch.

• Consistent portraits and product imagery built from real photographs
• Masked edits, companion images and transparent cutouts for layout
• Brand artwork using saved logos, colors and references

A model's answer is one step in a longer job. I build the interface, saved records and rules around it: where a person reviews a suggestion, how edits are kept, what gets exported, who can see it and how someone picks the work up next week.

I work through the steps on either side of the AI response. A suggestion may need approval before it changes a record. An illustration needs to stay with its story. A saved draft needs enough history for its author to return and understand the choices already made.

• Guided workflows, document tools and conversational interfaces
• Review steps, revision history and exports
• Private workspaces with owner, editor and viewer roles

I turn recurring jobs into assistants that act inside defined permissions: which requests they accept, which actions they may take and when they stop and hand the case to a person. The routine cases get handled, and the unusual ones arrive with their context attached.

I separate the rules for taking action from the language used to explain it. An assistant can write a friendly reply or summarize a morning's checks while the application controls account access, permitted actions and when a failed task needs a person's attention.

• Email assistants tied to account permissions
• Service monitoring with rechecks and a short list of allowed repairs
• Plain-language reports of what ran, what failed and what was tried

I can fit a local model to the hardware, documents and response time a job needs, then build the chat, task presets, queue and progress view around it. An older computer can have another working life if the application is designed for what it can handle.

Memory and waiting time shape the design. I work on keeping a model ready, selecting the relevant document passages and showing whether a request is queued, reading or answering. Running inference locally can also keep business files out of an external AI service during that work.

• Local inference with llama.cpp on existing hardware
• Private document chat that reads long files in passes
• Local models for data cleanup and in-game roles

I can bring maintenance questions into focus before asking a model to answer them. The Maximo release, the asset and its work history all matter. So does knowing whether the evidence is a general manual or a record from the system your team actually uses.

A planner, system administrator and technician may ask about the same work order for different reasons. I design tools that account for that context, explain the source behind an answer and leave the maintenance decision with the people responsible for the equipment.

• Maximo and reliability explanations with sources and release context
• Maintenance spreadsheet cleanup that protects identifiers
• Read-only connections that bring approved Maximo data into a conversation