Thomas Miksch
Senior AI Full Stack Engineer
7+ years delivering production AI systems across healthcare, real estate, and fintech. Specializing in agentic LLM pipelines, multi-modal AI, and real-time WebRTC/LiveKit infrastructure.
About Me
A passionate engineer driven by the intersection of AI and real-world impact
I'm a Senior AI Full Stack Engineer with over 7 years of experience delivering production AI systems across healthcare, real estate, and fintech industries.
At Theoria Medical, I led end-to-end development of AI-driven clinical tools—integrating IoT medical devices, voice/video AI agents, and AWS Bedrock-backed LLM services—that directly improved physician workflows and patient outcomes.
My expertise spans agentic LLM pipelines, multi-modal AI, real-time WebRTC/LiveKit infrastructure, and Physical AI for the medical domain. I have a consistent track record of scaling platforms from zero to enterprise with focus on latency, reliability, and regulatory compliance (HIPAA).
7+ Years
Professional Experience
B.S. Computer Science
University of Iowa, 2014-2018
HIPAA Compliant
Healthcare AI Systems
Technical Skills
A comprehensive toolkit for building production-grade AI systems
AI Agents & LLMs
ML Engineering
Data Science
Physical AI & IoT
Voice, Video & Avatar AI
Languages
Frontend
Backend & APIs
Data & Streaming
Cloud & DevOps
Compliance
Tooling
Work Experience
Building impactful products at the intersection of AI and enterprise software
Led end-to-end development of AI-driven clinical tools spanning multi-agent LLM systems, real-time voice/video/avatar AI, and edge-to-cloud Physical AI in a regulated, HIPAA-compliant environment.
Agentic AI & Multi-Cloud LLM Infrastructure
- ▹Designed a multi-agent clinical decision support system on AWS Bedrock (triage, medication review, ICD-10, SOAP-note agents) reaching 90%+ task completion across 400+ physician accounts
- ▹Built a production RAG pipeline over 2M+ clinical documents (LangChain, LlamaIndex, Pinecone) reducing hallucination rate ~65% vs. baseline
- ▹Implemented self-correcting agent loops with LLM-as-judge validation, cutting unsafe-output rate below 5%
- ▹Led a multi-cloud LLM strategy with Azure OpenAI failover and GCP Vertex AI / Gemini Enterprise for grounded retrieval via Google ADK & Agent Builder
- ▹Fine-tuned clinical NER models for PHI detection with Azure ML and AI Foundry prompt-flow orchestration
Real-Time Voice, Video & Avatar AI
- ▹Architected a LiveKit SFU telemedicine platform for 10,000+ concurrent sessions at <800ms round-trip (P95) via Deepgram STT, LLM chain, and ElevenLabs TTS
- ▹Developed auto-SOAP-note generation from live transcription with FHIR R4 write-back to Epic, cutting documentation from 10 minutes to 90 seconds
- ▹Engineered a real-time AI avatar agent (LiveKit + Tavus & Hedra, latent-diffusion fallback) autonomously handling patient intake and follow-ups
- ▹Built a video intelligence layer on WebRTC with GPT-4V for wound assessment and screening, surfacing findings within 3 seconds
Physical AI, IoT & Platform
- ▹Designed an edge-to-cloud Physical AI platform for 500+ bedside IoT devices (MQTT, AWS IoT Core, Kafka, Spark; 50M+ datapoints/day)
- ▹Deployed TFLite arrhythmia models on ARM Cortex-M4 at 35ms with a per-patient digital twin enabling 20+ minute earlier deterioration alerts
- ▹Built the Next.js 14 physician dashboard on an async FastAPI service with Redis LLM caching, sustaining 8,000 req/min at P99 <180ms
- ▹Enforced HIPAA & SOC 2 Type II: Presidio PHI redaction, Bedrock private VPC, KMS field-level encryption, TLS 1.3, and X.509 device certs
Owned end-to-end deal flow pipeline for a real estate investment platform serving 5,000+ accredited investors.
Key Achievements
- ▹Built real-time collaboration features reducing message latency from 900ms to <80ms
- ▹Architected analytics pipeline processing 500K+ property transactions weekly
- ▹Handled $50M+ in monthly transaction volume via DocuSign, Plaid, and Stripe integrations
- ▹Improved Lighthouse score from 54 to 91 via SSR optimization
- ▹Built semantic search system cutting analyst due diligence time by 60% (pre-LangChain era)
Built responsive investment dashboards and microservices for a women-focused fintech investment platform.
Key Achievements
- ▹Increased funnel completion rate by 22% through UX refactors and A/B testing
- ▹Developed Node.js microservices for portfolio rebalancing with brokerage API integration
- ▹Raised test coverage from 48% to 87%, cutting regression bugs by 40%
- ▹Built internal tooling for Customer Success triaging 2,000+ accounts/week automatically
Featured Projects
Production AI systems delivering real business impact
Clinical AI Voice Co-Pilot
Theoria Medical - 2024-2026
Real-time voice AI assistant for physicians during consultations with automatic SOAP note generation and FHIR integration.
- ▹LiveKit voice session -> Deepgram streaming STT -> GPT-4o/Claude function-calling chain
- ▹Structured SOAP note -> FHIR write-back with <800ms E2E latency at P95
- ▹Multi-modal context: lab results, radiology text, wearable vitals fused into single LLM context
<800ms
Latency
400+
Users
IoT Medical Telemetry & Edge AI
Theoria Medical - 2023
Edge-to-cloud IoT telemetry platform for real-time patient monitoring with on-device AI inference.
- ▹MQTT -> AWS IoT Core -> Kafka -> real-time Spark Streaming for 500+ devices
- ▹TFLite arrhythmia model on-device at 35ms inference time
- ▹AI-ranked alert prioritization reducing ICU nurse alert fatigue by 34%
500+
Devices
34%
Alert Reduction
RAG Deal Intelligence Engine
Agora - 2022
Semantic search system over deal documents for real estate investment analysis, built before the LLM revolution.
- ▹Semantic search over 10K+ deal documents using sentence-transformers + FAISS
- ▹Built before LangChain (Oct 2022) and GPT-3.5 (Nov 2022) were available
- ▹Cut analyst due diligence time by 60%
10K+
Documents
60%
Time Saved
AI Trend Alignment
Expertise aligned with the cutting edge of AI development
Agentic AI
Multi-step agents with tool use, planning & self-correction-built at Theoria across clinical workflows
Multi-Modal AI
Text + audio + image + time-series fusion in clinical LLM pipelines (GPT-4V, Deepgram, FHIR)
Domain-Specific AI
Medical-grade LLMs, HIPAA guardrails, fine-tuning on clinical notes, FHIR integration
Edge AI & Hardware
TFLite on ARM microcontrollers, offline inference, edge-to-cloud IoT architecture
Get in Touch
I'm always open to discussing new opportunities, interesting projects, or just having a conversation about AI and technology.