Cookies on this site

We would like to use cookies to understand how the site is used and to measure our campaigns. Nothing is set until you choose, and you can change your mind at any time from “Cookie settings” at the foot of every page. Privacy policy

Contact UsContact Us

AI agents · Agentic workflows · On-device AI

AI agents that do real work inside your business.

We design, build and run agents and agentic workflows on Microsoft 365, Azure and local models — document intake, label comparison, customer-data sync, research, procurement triage — with a safety gateway on every prompt and a data foundation you can trust.

Book a strategy callSee our AI work
  • Copilot Studio agents
  • Azure AI Foundry
  • RAG on governed data
  • Claude & Azure OpenAI
  • Local models (Ollama · NVIDIA GB10)
  • Prompt-injection defence

Hybrid AI, routed by policy

live

  • On-device

    Classify, redact, extract — sensitive data never leaves the laptop or edge box.

    30B local model · 11× faster than 120B

  • Your infrastructure

    Retrieval over governed SharePoint and database content. Grounded answers, cited.

    RAG · Azure AI Search · Purview labels

  • Cloud

    Scale-out agents, orchestration and monitoring when the task needs it.

    Azure OpenAI · Copilot Studio · Foundry

Every prompt and answer logged to Princeton Sentinel for audit

2,000 → 50

procurement listings a day, triaged by a local model

15×

faster than a 120B cloud-class model, same verdicts

$0 / prompt

local prompt-injection classifier, no API calls

100+

active clients across NJ, NY and PA

AI agents & agentic workflows

Agents that run inside the tools you already own

Not a chatbot bolted onto a website. We build agents that watch inboxes and SharePoint folders, read documents, call your systems, make bounded decisions and hand the rest to a person — with every step logged.

Explore agent services

Anatomy of one of our agentic workflows

Certificate-of-analysis intake · in production since April 2026

  1. Trigger

    Watch

    Outlook inbox, SharePoint folder or a schedule. New COA PDF arrives.

    Power Automate · Graph API

  2. Agent

    Reason

    Copilot Studio agent classifies the document and plans the extraction.

    Copilot Studio · Azure OpenAI

  3. Tools

    Act

    Extract 14 fields, validate against the schema, write the Excel row, file the PDF.

    AI Builder · Document Intelligence · SharePoint

  4. Safety gate

    Check

    Prompt-injection and data-poisoning scan; low-confidence fields routed to a reviewer.

    Safety gateway · Purview labels · human-in-the-loop

  5. Outcome

    Deliver

    Report in Teams, link to the filed document, exceptions queued for a person.

    Teams · Excel · email

↳ Telemetry: every prompt, tool call and verdict streams to Princeton Sentinel — usage, errors, RAI flags and cost per run

Built with: Copilot Studio · Power Automate · AI Builder · Azure Functions · SharePoint · Purview

Agents we have shipped

Multiple Agents in Production across varying Industries

  • Food manufacturing

    Document Intake Agent

    Watches Outlook for certificates of analysis, extracts fields with AI Builder, validates, files to SharePoint and reports in Teams. Hardened against prompt injection in the PDFs themselves.

    • Copilot Studio
    • AI Builder
    • Power Automate

    Hours of daily filing → minutes

  • Food manufacturing

    Label Comparison Agent

    Compares two versions of a product label, highlights every change, annotates the PDF and returns an audit-ready diff. Rebuilt on Azure Functions when Copilot Studio hit its limits.

    • Azure Functions
    • Document Intelligence
    • Python

    Audit-ready diffs in one chat turn

  • Food manufacturing

    Customer Lifecycle Sync Agent

    Keeps customer records, approvals, order history and profile changes in step across Magento, SugarCRM, M3 and Bloomreach, and logs what needs a human.

    • Copilot Studio
    • Power Automate
    • REST connectors

    Four systems, one customer record

  • Any department

    Analytics & Reporting Agent

    Reads spreadsheets and documents in SharePoint, runs real Python analysis — not summaries — and returns charts, Word and Excel outputs from the chat.

    • Copilot Studio
    • Python
    • SharePoint

    Reports without analysts in the loop

  • Internal · food & beverage

    Regulatory Research Agent

    Ingests FDA guidance, approval letters and labeling requirements; answers with citations, compares documents, drafts submission summaries. Built on Claude.

    • Claude Sonnet
    • RAG
    • Azure

    Cited answers over regulated content

  • Public-sector sales

    Procurement Scout

    Logs in to five procurement portals daily, reads 2,000 listings, parses solicitation documents and short-lists 20–50 bids — every model call on a local 30B model.

    • Playwright
    • Ollama · Nemotron 30B
    • SQL Server

    2,000 rows → 50, no cloud API

coa-intake-agent · run #4,812 · 06:14 ET
06:14:02trigger3 new attachments in Quality@ inbox (pdf ×3)
06:14:03agentclassified: certificate_of_analysis ×3 · plan: extract → validate → file
06:14:05toolsAI Builder model coa-v7 → 14 fields · confidence 0.97 / 0.96 / 0.91
06:14:05safetyprompt-injection scan: clean ×3 · poisoning heuristics: clean · labels: Internal
06:14:06reviewbatch #3 field "lot_no" 0.71 → routed to reviewer (jmartin)
06:14:07deliverfiled /Quality/COA/2026-10 · Excel row appended ×3 · Teams summary posted
06:14:07telemetryrun logged to Sentinel · 4 tool calls · $0.014 · 5.2 s

How we grow an agent

From copilot to autonomous — one rung at a time

Every successful deployment we have studied or shipped started in tool mode and earned autonomy. We move clients up this ladder only when the telemetry says the rung below is reliable.

  1. Copilot / tool mode

    A person asks, the agent answers or drafts. Zero autonomy, immediate value.

  2. Task agents

    One bounded job end-to-end: intake a document, compare two labels, sync a record.

  3. Agentic workflows

    Multi-step plans across systems with checkpoints, retries and a reviewer queue.

  4. Autonomous with guardrails

    Runs on a schedule, decides within policy, escalates by exception, reports to Sentinel.

Built on

  • Microsoft Copilot Studio
  • Azure AI Foundry
  • Azure OpenAI
  • AI Builder
  • Power Automate
  • Microsoft Graph
  • Dataverse
  • Azure Functions
  • Claude (Anthropic)
  • Ollama
  • NVIDIA GB10
  • Microsoft Purview
  • Azure AI Search

Four practice areas

Everything an agent needs to be trusted in production

Agents are the visible part. These four practices are what make them safe, accurate and affordable to run.

  • AI Agents & Copilots

    Design, build and run agents and agentic workflows on Copilot Studio, Azure and Claude — from first task agent to scheduled autonomy.

  • AI Safety & Governance

    A safety gateway on every prompt: injection and poisoning defence, Purview labels that travel with data, RAI telemetry in Sentinel.

  • Hybrid & Edge AI

    The right model in the right place — on-device, on your servers or in the cloud — routed by sensitivity, latency and cost.

  • Data Readiness for AI

    Posture reviews, labelling, retention and access clean-up so Copilot and agents only see what they should.

Hardware + AI

AI runs on hardware. We supply both.

On-device agents need AI-ready laptops. Local models need GPUs. Retrieval needs storage, and every agent needs a secure network to reach the systems it works in. Tech365, our value-added reseller division, sources, licenses and deploys the equipment our engineers design the AI for, so one plan covers the model and the metal.

Procure at Tech365.io →Apple for business →

  • Endpoints

    AI PCs and Copilot-ready laptops

    → on-device agents and Copilot

    Classification, redaction and extraction on the laptop, with no data leaving it. We size the fleet to the workloads, Tech365 supplies and images it, Princeton Service Center keeps it running.

    • Lenovo
    • Dell
    • HP
    • Apple
    • Microsoft Surface

  • Compute

    GPU workstations and desktop AI systems

    → local models and private RAG

    Our Procurement Scout runs 30B-parameter models on an NVIDIA GB10 desktop in our own office. The same box, or a GPU server, puts a private model next to your regulated data.

    • NVIDIA
    • ASUS
    • HPE
    • Lenovo

  • Infrastructure

    Servers, storage, networking and security

    → data platforms and safe agent access

    Agents call SharePoint, databases and ERPs over the network. Fortinet, Cisco and Meraki keep that path segmented; HPE and IBM storage hold the data lake underneath the models.

    • HPE
    • IBM
    • Cisco
    • Fortinet
    • Meraki
    • Veeam

Government and education buyers: Tech365 answers RFPs and holds the contracts; Princeton IT Services scopes the AI. One response, two divisions.

How the divisions work together

How we work

Prepare the data. Activate AI. Optimize and scale.

The same three-stage path we use with every client — so AI projects start on governed data and finish with measurable adoption.

  1. Prepare the data

    Visibility and trust: where data lives, who can reach it, what is overshared. Sentinel posture dashboard, labels, retention.

  2. Activate AI

    Workflow-embedded agents, not generic chat: analytics, document intelligence, automation — built in Copilot Studio and Azure.

  3. Optimize & scale

    RAG for accuracy, telemetry for adoption and ROI, safety gates for every prompt. Hybrid routing to keep costs down.

Work

Recent AI projects

Client names withheld; details available on a call.

See all case studies
  • In production

    Hours → minutes

    Food manufacturing

    AI Consulting

    Document intake agent for a national food cooperative

    A Copilot Studio agent reads certificates of analysis as they arrive in the quality team's inbox, extracts 14 fields with AI Builder, checks them, files the PDF in SharePoint and posts a summary in Teams. Fields it is unsure about go to a person instead of being guessed.

    • Hours of daily filing down to minutes
    • 14 fields per document, validated
  • Internal project

    2,000 → 50 a day

    Public-sector sales

    AI Consulting

    Procurement triage on a local 30B model

    Logs in to five government procurement portals each morning, reads about 2,000 listings, parses the bid documents and emails a short list of the 20 to 50 worth quoting. Every model call runs on one desktop AI system in our office.

    • 2,000 listings down to about 50 a day
    • 11× faster than a 120B model, same verdicts
  • Product

    Every prompt screened

    Any industry on Microsoft 365

    Microsoft

    AI activity and data posture for Microsoft 365

    Sentinel watches Copilot and Copilot Studio agents in a Microsoft 365 tenant. It logs prompts and tool calls, scores each agent's risk, flags oversharing and stale access, and screens prompts for injection attempts with a classifier that runs locally at no per-call cost.

    • Every agent prompt screened
    • Runs inside the customer's own Azure tenant

Customers

Who we work with

Partnerships

Built on the platforms you already use

All partnerships →

Products

Software we run for you

Built for our own AI practice, now available to clients.

  • Princeton Sentinel overview dashboard

    Princeton Sentinel

    AI safety and data posture for Microsoft 365

    Copilot Studio agent telemetry, prompt-injection classification, oversharing and stale-access detection, executive posture trends. Entra ID sign-in, runs in your Azure tenant.

    See Sentinel →
  • Business Card Intelligence lead pipeline

    Business Card Intelligence

    Event lead capture with AI follow-up

    Scan cards, PDFs or QR codes, get deduplicated leads, AI-drafted follow-ups in your own Outlook, reminders and company news. Mobile app in beta.

    See Lead Capture →

Services

Nine practices, one account team

AI leads. Underneath it sit twenty years of infrastructure, database and Microsoft work, and managed IT as your Managed Service Provider (MSP). Each practice has its own page and case studies.

Talk to an expert

Insights

From the engineering team

118 articles this year on Copilot, Purview, Azure and hybrid AI.

All articles

One group, seven brands

Everything from the AI to the hardware it runs on, under one account team

Meet the divisions

Not sure where to start? Book a 30-minute AI readiness call.

We will look at one workflow and tell you what is automatable, what is not, and what the data needs first.

Book a strategy call