Founding AI resource
Standing up an enterprise AI function
Leading AI engineering from the ground up across a national network of behavioral health centers — infrastructure, governance priorities, and delivery.
// Joseph Bisaccia · AI Engineering Leader
I build AI systems —
and the organizations
that trust them.
I lead enterprise AI adoption end to end: secure infrastructure, agentic workflows, and the training programs that turn skeptical teams into confident AI operators.
Focus
Enterprise AI Governance, Security, Infrastructure
Currently
Lead AI Engineer, Behavior Frontiers — building the AI function from the ground up
Open to
AI engineering leadership, strategic collaborations & speaking
Scroll
// 01—THE SYSTEMS
A short tour of the work.
Expertise
Production AI systems for regulated and enterprise environments.
- [00]Enterprise AI Governance
- [01]AI Security
- [02]Enterprise AI Infrastructure
- [03]HIPAA-Compliant AI Systems
- [04]Agentic Workflow Development
- [05]Retrieval-Augmented Generation (RAG)
- [06]Company-Wide AI Enablement
- [07]Technical Program Leadership
- [08]Change Management & Adoption
- [09]LLM Evaluation
// 02—ORGANIZATION
Building the systems — and bringing the organization along.
01
0%
Faster response times from the RAG assistant
02
0%
Reduction in manual operational work via automation
03
0%
Reduction in setup time across workflows
Leadership & enablement — a snapshot
Governance & security
Compliance-oriented implementation
Setting governance-oriented AI priorities with clinical, operations, and department stakeholders so enterprise workflows stay auditable and compliant.
Training enablement
Change management for distributed teams
Leading staff training and change management so distributed teams adopt AI tools responsibly — building on a decade of instructional design and district-wide rollouts.
Frontier model work
Model training & evaluation
Ongoing contract work with Handshake AI, Outlier AI, and Mercor: dataset curation, reward-metric evaluation, and RLHF preference workflows.
// 03—PERSPECTIVE
“Enterprise AI success is an organizationalchallenge, not an engineering one.The model is the easy part.”
The hard part is governance, trust, and adoption — the slow work of building something a regulated organization can actually stand behind.
The leaders who win the next decade won’t just ship models. They’ll build secure systems and bring entire organizations along — training the skeptics, designing the guardrails, and turning AI from a pilot deck into infrastructure.
I work at exactly that intersection: hands-on engineering depth, paired with the organizational enablement that makes the engineering matter.
// 04—WORK
Nothing ships until it passes a gate.
Open-source evaluation and governance harnesses — runnable, tested, and CI-checked. Every dataset is synthetic and every repository documents its limits.
~/projects/rag-gate▍
github.com/jbisaccia-9
jbisaccia-9/rag-gate
Retrieval gate that only serves at recall@3 ≥ 0.90. Baseline caught at 0.83, fixed to 1.00 on the current small synthetic set.
RAGRetrieval evalCI gate
View repository ~/projects/kappa-gate▍
github.com/jbisaccia-9
jbisaccia-9/kappa-gate
LLM-as-judge calibration on Cohen's kappa — a judge is trusted only at kappa ≥ 0.70 and agreement ≥ 0.85. The mock judge is refused.
LLM-as-judgeCalibration
View repository ~/projects/perm-gate▍
github.com/jbisaccia-9
jbisaccia-9/perm-gate
Prompt-layer guards versus permission-layer enforcement: on the synthetic set, prompt mode leaked 4/5 and permission mode 0/5.
AuthorizationSecurity
View repository ~/projects/phi-gate▍
github.com/jbisaccia-9
jbisaccia-9/phi-gate
Regex-tier PHI-shaped redaction gate — recall 1.00, precision 0.95 on the current synthetic corpus. Free-text names and addresses out of scope.
PHIRedaction
View repository Professional engagements
Behavior Frontiers · Healthcare
Founding the AI function in regulated healthcare
Building HIPAA-compliant AI infrastructure, RAG pipelines, and agentic workflows for clinical and operational teams across a national network of autism and behavioral health centers.
See detailsCapital Energy · Solar
Customer-facing RAG chatbot with agentic logic
Designed and deployed an inbound-query assistant that reduced response times by 40% and increased self-service adoption.
See detailsCapital Energy · Sales operations
Outbound lead reactivation agent
Voice and messaging agent built with Voiceflow, Twilio, and ElevenLabs to automate prospect engagement and accelerate pipeline.
See detailsOpen source
Seven gates, public and runnable.
Retrieval, judge calibration, permissions, PHI redaction, ROI assumptions, targeting pipelines, and order validation — each with a threshold that has to be earned before anything ships. Client and employer work stays private.
github.com
jbisaccia-9
// 05—CONTACT