JP Siler. Résumé

Hi, I’m JP.

Software engineer · Applied AI

Useful AI. Thoughtfully built.

I turn complex problems into practical AI products and dependable software.

Selected work Let’s talk

Houston, TXRemote & open to relocation

SELECTED WORK

Ideas, put to work.

Two AI products. Built, shipped, and used by paying customers.

01

REAL-TIME VOICE · 2024–2026

Voice/Vision AI Platform

AI phone agents that answer calls, screen applicants, and book appointments—with company knowledge and human review built in.

  • OpenAI Realtime
  • Twilio
  • WebSockets
  • WebRTC
  • Python
  • Pinecone
  • Stripe
Live version coming soon

Replay · AI-generated voices · sample data

Ridgeview Packaging

Screening call · Maintenance Technician (Nights)

Supervisor agent 00:00

    Problem I solved

    Redaction you can rely on.

    A single missed age or family detail in a hiring transcript is one too many, and model-only redaction isn’t deterministic. I paired LLM redaction and a verification pass with a fixed lexicon: lexicon terms are always removed, redacted spans are muted in the recording, and if redaction fails, the transcript, audio, and PDF are withheld rather than released.

    How a call flows
    1. CallerPhone · Twilio Media Streams · MMS photos
    2. OpenAI RealtimeWebSockets, live audio
    3. Supervisor agentSilent handoff to specialists · knowledge search
    4. Queue workersTranscript · summary · bias filter · PDF
    5. Human reviewSigned URLs · approve and finalize

    Tenant-scoped data · audit-logged, read-only support access · idempotent usage ledger · Stripe

    How it’s built

    A voice AI platform serving paying small-business customers. I designed, built, and operated the product, from real-time conversations to billing and customer support.

    Real-time conversations

    Twilio Media Streams connect to OpenAI’s Realtime API over WebSockets. A supervisor agent hands live calls to specialist agents, while retrieval over company documents provides context during the conversation.

    From call to action

    Agents screen applicants, answer calls, book appointments, and place outbound calls. The platform includes outbound batch campaigns, automated phone provisioning, and a browser-based dialer with dual-channel recording, transcription, and call history.

    The work after the call

    Queue-driven Python workers handle transcripts, Spanish-to-English translation, AI summaries, and branded PDF reports delivered through signed URLs. Staff review and approve AI-drafted call reports before they are finalized, with optional e-signature and SMS linked to the call.

    Safeguards built into the workflow

    A bias filter combines LLM redaction and verification with a deterministic lexicon to redact protected-characteristic disclosures from candidate transcripts, recordings, and PDFs. Opt-outs are audit logged. Company data is tenant scoped, and support access is audit logged and read-only by default.

    A product that can operate

    A prepaid credit wallet, idempotent usage ledger, and Stripe subscriptions connect usage to billing. Customer onboarding and operational support are part of the engineering work.

    02

    KNOWLEDGE & WORKFLOWS · 2026

    AI Form/Workflow Platform

    Custom AI assistants grounded in each business’s own data, with structured outputs, tenant isolation, and accountable automation.

    • Retrieval-augmented generation
    • Structured outputs
    • Multi-tenant architecture
    • Metered billing
    Live version coming soon

    Message

    Pick a sample message.

    Form

    Structured output

    {}

    Problem I solved

    From 6–8 seconds to 2–3.

    Ask AI answers were taking six to eight seconds. Profiling showed the time going into embedding and vector search, not the model. When a business’s documents fit in the prompt, the assistant now skips retrieval entirely, and the first text appears in about two to three seconds.

    How an answer is produced
    1. Form or chatVisitor question or one-sentence fill
    2. AssistantOwner notes + knowledge base
    3. Fits in prompt?Yes: direct context · No: retrieval (Pinecone)
    4. Structured outputSchema check · bounded retry
    5. WorkflowApprovals · e-sign · notifications

    Tenant data isolation · audit-logged agents · risk classification at intake · metered billing

    How it’s built

    A multi-tenant platform serving paying customers, where businesses configure custom AI assistants over their own data. I designed, built, and operated the product.

    Grounded in business knowledge

    Retrieval-augmented generation connects assistants to uploaded knowledge bases. Deterministic computation operates over customer reference data and is read-only by design.

    Predictable output

    Schema-validated structured LLM outputs and bounded retry give downstream workflows an explicit output contract.

    Boundaries and accountability

    Tenant data isolation separates business data. Autonomous agents are audit logged, and intake includes risk classification for regulated data.

    Beyond the application

    Metered usage billing supports commercial operation. Onboarding training and instructional videos help customers put their assistants to work.

    Earlier projects
    Knowledge search for constructionConversational search over proprietary field-labor data · 2022–2023

    CLIENT ENGAGEMENT

    Retrieval-based conversational AI built around a construction company’s proprietary field-labor database. The work connected company-specific information with a conversational way to find it, powering customer-facing search.

    • Conversational AI
    • Retrieval
    • Customer-facing search
    Skilled-trades hiring platformWork-sample portfolios and video calls · 2018–2021

    INDEPENDENT PRODUCT

    A hiring platform for skilled trades, bringing evidence of hands-on work into the hiring process. Work-sample portfolios and embedded video calls let employers see a candidate’s practical experience and speak with them.

    • Hiring platform
    • Work-sample portfolios
    • Embedded video

    A LITTLE ABOUT ME

    An engineer’s toolkit.
    An operator’s mindset.

    I design, build, and operate AI products, alongside my work on public-sector applications.

    Before software, I spent fifteen years in manufacturing—programming automation, solving operational problems, and leading a maintenance team. It taught me to build for the people using the system, and for what happens after launch.

    B.S., Industrial EngineeringTexas Tech University

    Experience, in brief.

    Full résumé
    Independent products & consultingDeveloper & product builder 2018–Present

    Commercial software development alongside full-time employment. Applied AI products and client work since 2022, covering the full path from product design and engineering to customer onboarding and operation.

    Texas Higher Education Coordinating BoardApplications Developer 2021–Present

    Full-stack development for Texas OnCourse and its successors using Node.js, Vue.js, and Python. Production workloads on Azure Kubernetes Service, plus core work on the AWS-to-Azure migration and MongoDB-to-Cosmos DB migration.

    Joined via InfoStride in November 2021; hired directly in May 2022.

    Anagh TechnologiesContract Developer Apr–Jul 2021

    Front-end development and Salesforce integration for Coach and Kate Spade e-commerce sites.

    Owens CorningMaintenance Supervisor 2006–2021

    Fifteen years in technical and leadership roles: PLC programming for robotics and servo control, plant projects, Lean Six Sigma events, and supervision of a 15-person maintenance department.

    My technical toolkit

    Build

    JavaScript, Node.js, Python, React, Vue.js, WebRTC, WebSockets

    Intelligence

    OpenAI, Anthropic Claude, AWS Bedrock, LangChain, CrewAI, RAG, voice and vision AI

    Infrastructure

    Azure, AWS, Kubernetes, PostgreSQL, MongoDB, Cosmos DB, Pinecone

    Trust & integration

    Tenant isolation, audit logging, human review, bias redaction, Twilio, Stripe, Salesforce

    LET’S CONNECT

    Something in mind?
    Let’s build it.

    I’d love to hear about your next project or opportunity.

    The product code is private. I’m happy to walk through it on a call.