22 projects across the range, from frontier-model research teams to a robot on a USDA line. 18 were built end to end; the rest architected, integrated, or led. The tag on each card says which, because honesty is part of the work.
All projects
BuiltEnterprise Mobility
SaaS-Hermes
A zero-trust agentic research workspace
Six specialist research agents work alongside a strategy team on Azure Container Apps, each with its own persona, methodology, citation rules, and cross-referral handoffs. Shared team memory compounds over time; a curated RAG index grounds every answer; frontier models run entirely in-tenant. Zero external LLM calls.
6 specialist agents
17 containerized services
19 lint-enforced invariants
0 external LLM calls
Azure AI Foundry · Container Apps · Azure AI Search · MCP · WebRTC · FastAPI
ArchitectedEnterprise Mobility
Private Enterprise GenAI Platform
The in-tenant control plane every solution plugs into
A fully in-tenant Azure AI platform where no inference or data ever leaves the company tenant. One governed, SSO-gated workspace delivering multi-model chat, agentic RAG, voice AI, and document intelligence, plus the reusable blueprints that turn one-off builds into a repeatable platform.
An AI-designed food-production robot, running daily on a USDA line
Designed, built, and deployed the first production unit into a USDA-inspected co-packer. AI generated the CAD for custom nozzles, designed the circuits, and helped write the firmware. Real hardware: a custom KiCad PCB (two fabricated revisions) with an ESP32-S3 and GRBL motion control, OTA updates, and a cloud fleet-management backend. Proof that agent-orchestrated AI solves physical problems, not just digital ones.
1st production unit deployed
running daily since 2026
2 provisional patents
AI-generated CAD + firmware
ESP32-S3 · KiCad · GRBL · OTA firmware · Google Cloud Run · Supabase
A self-hosted platform of 14 named AI agents (CEO advisor, operations, sales, marketing, finance, support, and more) that run Chicken Butt Jerky's business functions, with cross-agent long-term memory. The proof of the conductor model lives in the git log: agent personas author the majority of commits, with a human architect and reviewer.
Document intelligence that collapses weeks into hours
A 7-stage pipeline that scrapes a gated external portal, enriches every document with AI summaries and extracted action items, and serves it all as searchable knowledge, compressing weeks of manual review into a 2-hour auto-refresh cycle.
A pilot agent that lets support reps retrieve data and answer customer questions in natural language, surfacing answers in the conversation instead of hunting across systems. Direct proof of conversational AI tied to a measurable business outcome.
40% shorter calls
+23% accuracy
+17% CSAT
98% response accuracy in pilot
RAG · LLM retrieval · Conversational UX
BuiltEnterprise Mobility
Cost-Cascade Model Router
Frontier capability at a fraction of the spend
A token-efficient router that opens every request on a cheap model and escalates to stronger, frontier-class models only when the task demands it, driving spend down without capping capability.
Searches public sources and scrapes gated ones into a governed, trust-weighted RAG knowledge base with per-source allow/deny-listing, runs the analysis, and auto-generates decks, charts, podcasts, and short videos, with real-time voice plus STT/TTS. Delivered on OpenWebUI.
multi-agent, solo-built
trust-weighted RAG
auto-generated media
real-time voice
OpenWebUI · Azure AI Search · STT/TTS · headless scraping
BuiltEnterprise Mobility
AI Coach Layer
Adoption engineered into the product
Three cooperating agents built on top of the research ecosystem: an avatar-and-voice onboarding trainer with ~12 lessons that spotlight the live UI, a passive on-demand coach, and a prompt-coach that watches a user's prompts to other agents and rewrites them for better results.
3 cooperating agents
~12 guided lessons
live prompt rewriting
Azure VoiceLive · WebRTC · avatar + TTS
BuiltEnterprise Mobility
CarIntellect
Multi-model research with built-in conflict detection
A queue-driven pipeline that discovers and scrapes sources, then runs three independent AI analyses with explicit conflict detection, producing a citation-tracked, 14-section executive dossier with confidence scoring.
Calls that bridge a realtime voice model to telephony, hold a natural 10-state conversation, and auto-chain a notification when a vehicle is ready, with custom audio transcoding between telephony and model formats.
Proactively monitors knowledge bases and briefs users out loud through a lip-synced avatar, with a fast-path design that answers in roughly two seconds and leader-elected monitoring across many indexes.
Containerized Model Context Protocol servers delivering ~45 tools (knowledge retrieval, voice, SQL analytics, live web scraping, presentation generation), consumable by any MCP-compatible agent, plus a standard factory lifecycle for building, deploying, and operating domain MCP servers with quality and telemetry gates.
~45 tools
containerized servers
standard build/deploy lifecycle
FastMCP · Azure Container Apps · Azure AI Search
ArchitectedEnterprise Mobility
Enterprise AI Chat
One workspace for the whole org
An Entra-gated workspace giving teams frontier models, knowledge-base search, voice calling, slide generation, diagramming, and analytics through natural language, all in-tenant. The hub every other solution plugs into.
A drag-and-drop builder where business users design multi-step AI pipelines and publish any flow as an MCP server, instantly turning a visual workflow into a callable tool.
A DFRobot ESP32-S3 AI camera turned into a hands-free voice-and-vision terminal: on-device audio capture with voice-activity detection streaming over a private Tailscale mesh to a Whisper-to-LLM-to-TTS bridge, with a camera path wired for a planned HACCP food-safety inspector.
Hebrew text to Chatterbox TTS to RVC v2 conversion, running entirely on an M1 Mac. Rigorously evaluated: CER ~3% intelligibility and an ECAPA voice-match of 0.664, about the 48th percentile of the speaker's own real held-out recordings.
100% on-device
CER ~3%
ECAPA match 0.664
Chatterbox TTS · RVC v2 · Apple Silicon
BuiltChicken Butt Jerky · Meat Bot
Breakout
AI-enriched GTM intelligence in one command
Turns raw distributor and field reports into a geographic, ICP-scored sales dashboard, enriching stores with trade-area and demographic data.
one-command pipeline
ICP scoring
geo + demographic enrichment
Python · SQLite · geocoding + enrichment APIs
BuiltEnterprise Mobility
Cloud Infrastructure Transformation
Static estate to auto-scaling, auto-healing
Re-architected a static hub-and-spoke estate into a demand-based, auto-scaling, auto-healing model, retiring over-provisioned appliances for dynamic capacity.
80% faster incident response
45%+ lower opex
auto-healing
Terraform · Ansible · serverless functions
BuiltEnterprise Mobility
Reference Architectures & Guardrails
Governance by automation, not paperwork
The org's AI reference library: a zero-trust private-AI blueprint, an MCP factory pattern, an Agent Operations Center (LLMOps across runtime health, knowledge freshness, tool reliability, and change governance), and a platform-resilience blueprint. Guardrails that cannot regress: 19 hard invariants enforced by lint on every deploy, plus architecture decision records with revisit triggers.
4 reusable blueprints
19 lint-enforced invariants
targets 99.5% availability
LLMOps / AgentOps · IaC · policy-as-lint · ADRs
LedEnterprise Mobility
Innovation & Evaluation Framework
Turning ideas into funded, scored bets
A data-driven ideation engine that captures, scores, and prioritizes AI proposals on cost, KPIs, and business alignment, letting the department self-fund high-value bets. Paired with an enterprise API-security evaluation: a 140-requirement scoring matrix presented to senior leadership with a tested implementation plan.