Phil Maher · Conferences / Podcasts / Workshops
Give your audience a look inside the work: training a model for a specific visual world, building financial systems in Rust, and getting AI into daily operations. Phil brings the projects, the engineering decisions and the lessons your audience can use.
AI-generated examples from Phil’s Civitai releases. Explore all four training approaches below.
Recent speaking & interviews
Start with a recent appearance, then explore the full interview, chapters and transcript on Stockfilm.
Latest speaking engagement
Phil joins Nathan Labenz and Prakash Narayanan to discuss original film as AI training data, experimental face restoration, and TimeFly, his browser-based fly simulation.
Host Nathan Labenz has also interviewed, on The Cognitive Revolution
Hear Phil in conversation
Phil joins filmmaker Christian Taylor to talk about the family reels that started Stockfilm, rescuing fragile film, and the everyday details that bring a documentary’s past to life.
Start at 23:36: home movies and storytelling01 / Model training you can see
Phil’s ReplicantArmy releases explore a specific target: the look of 1947 Paris. The model descriptions document training on more than 2,200 archival images. These examples make the choices behind fine-tuning tangible.
REPLICANTARMY / CIVITAI MODEL 405790
A compact adaptation for a defined place, period and image style. This release is listed on SD2.1 768; its version notes describe testing across SD2.0 and SD2.1 base variants.
REPLICANTARMY / CIVITAI MODEL 395798
The same narrow visual subject, adapted for the SD1.5 ecosystem. A useful starting point for a discussion about choosing a base model and deciding what a specialized adapter needs to learn.
REPLICANTARMY / CIVITAI MODEL 404109
A checkpoint release brings a different packaging and deployment choice to the same creative problem. Phil can use the examples to explain when a full model artifact fits a workflow and what teams should evaluate before adopting it.
REPLICANTARMY / CIVITAI MODEL 392240
Another adaptation method in the series, published before the LoRA and checkpoint releases. Together, the four artifacts give a technical audience concrete examples to discuss compatibility, control and the cost of specialization.
Turn “make it look right” into a specific training objective, then examine the examples for what they capture and what they miss.
Discuss selection, consistency and the limits of archival reference material. Give the audience questions they can apply to their own data.
Use LoRA, checkpoint and hypernetwork releases to explain how model choices affect integration, iteration and creative control.
These are generated samples from the published model galleries. They illustrate the models’ visual output; they are not evidence of historical events. Training details are attributed to the release descriptions. See the ReplicantArmy model profile ↗
02 / Financial systems & HFT
For a technical conference or finance podcast, Phil can connect specialized model work to the architecture around it: data, evaluation, execution, telemetry and the decision to release a change.
Phil’s work includes fine-tuning models for specific financial and high-frequency trading client requirements. He can speak to the engineering questions behind that work: defining the task, evaluating on relevant data and fitting the model into an operational system.
Ferrix provides a detailed public architecture to explore alongside that experience. Its Rust design separates live execution from an evolutionary optimization loop, with telemetry and out-of-sample evaluation connecting the two.
The published design reserves the newest 5% of data as a permanent holdout and uses embargoed walk-forward folds. Accepted candidates trigger a controlled restart after active positions and processing have cleared.
Read the Ferrix repository overviewFerrix’s optimization loop searches strategy parameters. The Civitai releases adapt image models. These examples give audiences two distinct ways to discuss training, evaluation and deployment. Client sessions can cover engineering methods without identifying private clients or disclosing their models.
03 / The portfolio behind the conversation
Every project featured on AIAdvocate’s Work page has a place here. Pick the ones that fit your audience, from archive intelligence and creative AI to financial simulation and legal workflows.
Visual search / Agents / Production AI
An archival footage platform with visual embeddings, custom SigLIP2 classifiers, AI-assisted project briefs and rough cuts. Its workflows connect archive enrichment, review, licensing and delivery.
Audience takeaway: How to turn a large media collection into a useful product, with validation and human review at the points that need them.
Explore Stockfilm ↗Rust / HFT / Optimization
An event-driven Rust trading architecture for Solana token migrations. Live position management runs separately from a telemetry-driven optimizer, with holdout scoring and controlled promotion.
Audience takeaway: How to design a system that can search for improvements while maintaining clear ownership of live state and release decisions.
Explore the Ferrix architecture ↗
Replay / Simulation / Reproducibility
Historical Solana slot replay and Monte Carlo execution scenarios for examining inclusion, slippage and MEV exposure. Seeded runs and signed manifests make results easier to reproduce and compare.
Audience takeaway: How to test decisions against execution conditions, rather than judging a strategy only by its idealized output.
Explore ReplayState ↗
Legal workflows / Evidence / Access
A legal chronology workflow that keeps events connected to source evidence and context, with exports and access controls built around the work of reviewing a case.
Audience takeaway: Why useful automation preserves the link between a conclusion and the material a professional must verify.
Explore TimelineSystem ↗
Rust / Prediction markets / Risk
A Rust-based prediction market risk service focused on resolution rules, disputes and settlement delays. The product brings the mechanics behind a market into the decision-making process.
Audience takeaway: How domain-specific risk can shape a data product, its API and the questions users need answered.
Explore SettleRisk ↗
Generative AI / History / Interactive media
A place-and-year guessing game built around AI-generated historical panoramas. Its content pipeline and game modes turn a specialized visual concept into an experience people can explore.
Audience takeaway: How to move from generated images to a complete product, with content choices, interaction design and clear expectations about synthetic scenes.
Explore LatLongAgo ↗Project descriptions are based on Phil’s published portfolio. Read the full Work page ↗
Client delivery / Published case studies
For leadership audiences, Phil can connect implementation choices to the work they change: processing documents, finding internal answers, reviewing reports and deciding which projects to fund.
The published case study reports roughly 30 hours freed each week and an error rate reduced from 8% to under 2%.
Read the case study ↗A retrieval-assisted workflow made internal information easier to find, with answers linked to source documents.
Read the case study ↗The reporting case describes a shift from manual preparation to a shorter review cycle, with people checking the output.
Read the case study ↗The portfolio describes two vendor proposals declined and three projects selected for the first implementation phase.
Read the engagement overview ↗Figures above are outcomes reported in AIAdvocate’s published client case studies and engagement summaries.
04 / Repositories & developer takeaways
Phil’s ReplicantArmy repositories include a TypeScript market-data client, a Python research loader, an MCP integration and detailed financial-system architecture documents.
A technical session can start with an actual interface and follow the decisions behind it: normalized data, tool contracts, reproducibility and the boundary between an assistant and the system it controls.
Browse ReplicantArmy on GitHub ↗
import { UnifiedClient } from 'polykalshi';
const client = new UnifiedClient();
const markets = await client.search('bitcoin');One client for normalized Polymarket and Kalshi data. Example adapted from the repository’s documented interface.
Rust execution architecture, actor-managed positions, telemetry, evolutionary optimization and out-of-sample promotion gates.
Repository overview ↗Solana slot replay and seeded execution simulations, with signed manifests and execution-risk analysis.
Repository overview ↗A TypeScript library with a unified market-data interface, normalized types, pagination and retry handling for Polymarket and Kalshi.
Code & examples ↗An MCP integration for archive discovery and footage workflows, including search, clip details and rough-cut tools.
Integration & docs ↗A Python loader, sample manifest and citation guidance for a versioned visual research index.
Loader & research docs ↗Phil’s profile also includes forks and repositories used for community resource submissions. These links show the wider collection without attributing the upstream projects to him.
Repository directory reviewed October 8, 2026. Ferrix and BlockEnvy are presented through their published architecture overviews. The ReplicantArmy profile repository ↗ completes the public directory.
05 / Choose the conversation
Use these as starting points for an event brief. Each can become a conference talk, a technical discussion or a podcast episode shaped around the people you want to reach.
AI ENGINEERS / CREATIVE TECHNOLOGISTS
Use the Paris releases to unpack dataset choices, base models and LoRA, checkpoint and hypernetwork approaches.
People leave with: A clearer way to define a training objective and decide what evidence would show that an adaptation works.
Choose this topic ↗FINANCE / RUST / SYSTEMS ENGINEERING
Explore specialized client requirements and the Ferrix architecture: execution, telemetry, walk-forward evaluation and release gates.
People leave with: Concrete questions for assessing overfitting, operational state and the conditions for promoting an improvement.
Choose this topic ↗CTOS / PRODUCT TEAMS / AI LEADS
Compare approaches through internal knowledge retrieval, specialized image models and production integrations.
People leave with: A decision process grounded in their data, the task, evaluation needs and the cost of operating the solution.
Choose this topic ↗DEVELOPERS / PLATFORM & AUTOMATION TEAMS
Walk through MCP interfaces, archive workflows and the approval and escalation points described in Phil’s production work.
People leave with: A practical way to define tool boundaries, validate output and keep review connected to the underlying action.
Choose this topic ↗MEDIA / DESIGN / CREATIVE PRODUCT TEAMS
Connect the Civitai models, archive intelligence and LatLongAgo’s panoramas to the work of building an experience around AI output.
People leave with: Ways to plan a content pipeline, assess visual consistency and explain the limits of generated historical scenes.
Choose this topic ↗EXECUTIVES / OPERATIONS / BUSINESS OWNERS
Use published client cases to discuss workflow selection, vendor review, human oversight and adoption after the demo.
People leave with: A way to choose a first project, define a measurable result and plan for the people who will use it.
Choose this topic ↗A focused 30–60 minute session with visual examples, architecture and time for audience questions. Shape the depth around your program.
Follow one project in depth or connect the creative, financial and operational sides of applied AI. Give listeners specific work to explore afterward.
A longer session around a defined problem: model evaluation, a retrieval workflow, MCP tool design or an implementation plan.
06 / For producers & organizers
Phil Maher is an AI implementation consultant and software developer with more than 30 years of experience. Through AIAdvocate, he works across model adaptation, retrieval systems, automation and production architecture. His portfolio includes specialized image models published as ReplicantArmy, Rust financial systems and products for archival media, legal workflows and prediction markets. He speaks about the decisions behind the work: choosing a useful problem, evaluating the result and building a system people can operate.
Planning the appearance
Yes. Start with the workflow and the people using it, then use visual examples and client cases to explain the decisions. For an engineering audience, the same material can go deeper into interfaces, evaluation and architecture.
Yes. Include your audience and preferred depth in the invitation. Ferrix, ReplayState, BlockEnvy, SettleRisk and polykalshi provide concrete material for discussing financial systems, evaluation and execution risk. Client-specific models can be discussed through their engineering methods without naming private engagements.
Use the project links, four Civitai releases and repository directory above. The bio, photo and implementation playbook are available here, too. Mention the projects that interest you so the conversation can center on them.
Send the event or show name, audience, date or recording window, format and budget if you have one. Scope, availability, travel and fee arrangements are confirmed when discussing the invitation.
Start with your audience
Tell Phil what you’re planning and what you want people to learn. A conference room, a workshop or a podcast can each make use of a different part of the work.
Include your event or show, date, audience and preferred topic. Phil will respond to discuss the fit, availability and arrangements.
Use the general contact page ↗