Microsoft Agent Framework 1.0 · MCP · Azure AI Foundry

Production agentic AI
on .NET and Azure.

Microsoft Agent Framework. MCP. Azure AI Foundry. I build, harden, and ship the production systems that come after Microsoft's hello-world samples — for Microsoft-stack enterprises and Solution Partners.

Book a 30-Minute Strategy Call →
12yr
.NET production experience
MAF 1.0
Microsoft Agent Framework specialist
MCP
Server hardening + C# SDK
Azure
Foundry · AI Search · OpenAI
Hari Prakash
Hari Prakash
Founder & AI Systems Architect
12yr
.NET production experience
MAF 1.0
Microsoft Agent Framework specialist
MCP
Server hardening + C# SDK
Azure
Foundry · AI Search · OpenAI
Is This For You?
Built for Microsoft-stack engineering teams
shipping production agentic AI
If any of these sound familiar, we should talk.
🏢

Microsoft Solution Partners

You're delivering enterprise Microsoft Agent Framework or Foundry engagements and need a senior .NET specialist on the delivery team. Subcontracting through your partnership is welcome.

🔧

Mid-market SaaS on Microsoft stack

You've shipped Semantic Kernel 1.x to production and need to migrate to Microsoft Agent Framework 1.0 without behavioral regressions, or you're starting MAF/MCP fresh and need production-grade defaults.

🛠️

Engineering teams past the sample stage

You've cloned Microsoft's MAF samples and now hit the production gap: observability, MCP server hardening, evaluation in CI, OWASP-aligned guardrails. The boring-but-critical layer.

See It Running
This isn't theoretical.
Look at the code.
AgentOps.NET — my open-source production reference architecture for Microsoft Agent Framework + MCP + Azure AI Foundry. The libraries Microsoft's official samples skip: observability, MCP server hardening, evaluation in CI.
AgentOps.NET production reference architecture: Microsoft Agent Framework graph workflow with three specialist agents, hardened MCP server, hybrid retrieval over Azure AI Search and pgvector, and cross-cutting NuGet libraries for observability, hardening, and evaluation
Open source — github.com/pinusx-ai/agentops-dotnet
📊

OpenTelemetry Observability

Traces flowing to Application Insights and Langfuse. Cost-per-trace, tool-call spans, agent-handoff spans, PII redaction. Built-in dashboards.

🔐

MCP Server Hardening

Entra ID validation, scope-to-tool maps, signed tool descriptions, audit logging, approval gateway for destructive tools. OWASP-aligned defaults.

✅

Evaluation in CI

xUnit fixtures running Azure AI Evaluation SDK + RAGAS metrics from golden YAML. PR comments. Build fails on regression.

View the Repo →
The Process
From assessment to production
in weeks, not months
A structured process that de-risks the engagement. You see results before you commit to ongoing operations.

Architecture Assessment

I review your existing Semantic Kernel, MAF, or MCP implementation. Identify production gaps in observability, security, and evaluation. Deliver a concrete migration or hardening plan with risk analysis. Yours to keep regardless.

1–2 weeks Migration plan + risk analysis

Implementation

Migration from Semantic Kernel 1.x to MAF 1.0, MCP server hardening with Entra ID and OWASP defaults, OpenTelemetry observability wired to Application Insights and Langfuse, evaluation harness in CI. Tested in production.

2–8 weeks Production-grade system live

Senior Advisory

Available for ongoing senior advisory engagements: architecture review, MAF/MCP design decisions, production incident analysis. Hourly or retainer.

Ongoing Architecture + production support
Why Not DIY?
How specialists compare
The real cost of hiring an in-house engineer, engaging a large consultancy, or working with me directly.
Hire FTE Big Consulting Firm Work With Me
Timeline 3–6 months 3–4 months 4–6 weeks
True cost $200K+/yr loaded $250–$400/hr $150–$180/hr
What you get Ramp time + retention risk Account manager + juniors Direct work with the architect
Who builds it In-house team Account manager + juniors The specialist, directly
Ongoing support Internal team carries the cost $20K–$50K/mo retainer Hourly as-needed or retainer
You own the code Yes Depends on contract 100% — open-source by default
Pricing
Start with an assessment.
Engage as needed.
No retainer required. Fixed-scope assessments deliver concrete plans. Hourly advisory for ongoing implementation work.
MAF Migration Assessment
$8,000 – $12K
fixed scope · 2 weeks
Complete review of your Semantic Kernel codebase, migration plan to Microsoft Agent Framework 1.0, behavioral regression risk analysis.
  • Semantic Kernel 1.x audit
  • MAF 1.0 migration plan
  • Behavioral regression risk analysis
  • Code samples and patches for high-risk areas
  • Yours to keep — no obligation
Discuss Your Migration →
MCP Server Hardening Audit
$6,000 – $10K
fixed scope · 1–2 weeks
Review your MCP server against OWASP MCP guide. Deliver hardening plan with code patches for Entra ID, scopes, audit logging, approval gateway.
  • OWASP MCP threat model assessment
  • Entra ID + scope-to-tool authorization plan
  • Audit logging + approval gateway design
  • Reference patches in idiomatic .NET
  • Yours to keep — no obligation
Discuss Your MCP Server →
Senior Advisory & Implementation
$150 – $180
per hour · 40hr minimum
Production agentic AI architecture, MAF + MCP + Foundry implementation, evaluation harness, production support. Hourly or retainer.
  • MAF / MCP / Foundry implementation
  • OpenTelemetry + Langfuse observability
  • Azure AI Search hybrid retrieval
  • Evaluation harness in CI
  • Microsoft Partner subcontract welcome
Discuss Engagement →
Hari Prakash

Hari Prakash

Founder & AI Systems Architect · PinusX AI, LLC

12+ years of enterprise .NET engineering across aviation, healthcare, and SaaS. Shipped production agentic AI on Semantic Kernel + GPT-4o (zero-hallucination text-to-SQL on a live ERP) and built Kerno, a 14-stage multi-agent orchestrator running in production. Now building AgentOps.NET as the open-source production reference for Microsoft Agent Framework + MCP.

Now I work with Microsoft-stack enterprises and Microsoft Solution Partners as a senior specialist. You work directly with me — the engineer who designs and builds the architecture. No account managers, no juniors learning on your project.

Why a solo specialist, not a consultancy? I've shipped production agentic AI in .NET. I've debugged hallucinations in front of enterprise customers, hardened MCP servers against tool-poisoning, and migrated codebases from Semantic Kernel to Microsoft Agent Framework. That production depth — not slide decks — is what you're paying for.
View AgentOps.NET → LinkedIn → See Published Output →
Common Questions
Before you book the call
Can't my engineering team build this internally? +
They can — but Microsoft Agent Framework, MCP, and Azure AI Foundry have all GA'd within the last 90 days. Your team will spend 2–3 months learning the production gaps that this engagement closes in 2–4 weeks. Engineering hours are better spent on your product, not on rediscovering OpenTelemetry GenAI conventions or OWASP MCP defaults.
What if Microsoft Agent Framework changes? +
MAF 1.0 went GA on April 3, 2026. The public API is now stable with one-year support guarantees from Microsoft. Preview features (Process Framework, A2A 1.0) are explicitly quarantined in my work — only the GA surface goes into production engagements. When MAF adds features that supersede my libraries, I deprecate them transparently.
Do you work as a subcontractor through Microsoft Partners? +
Yes — that's a primary engagement model. PinusX AI, LLC is a Delaware-registered entity suitable for partner subcontracting. I work as a named specialist on your engagement, billing through PinusX, while your firm owns the client relationship. Microsoft Partner Network connections welcome.
What if you disappear? +
You own 100% of the code I write. AgentOps.NET libraries are MIT-licensed. Engagement code is delivered under your repo with zero proprietary dependencies — standard .NET, standard Azure, standard Microsoft Agent Framework. Any senior .NET engineer can maintain and extend the work using the documentation I provide.
How do you compare to Big Four / large consultancies on Foundry engagements? +
Different model. Large consultancies sell relationship-led delivery with senior partners selling and junior engineers building. I sell direct senior delivery — you talk to the architect who writes the code. For Foundry/MAF/MCP work where production depth matters more than scale, this often delivers more in 4 weeks than 4 months of consultancy time.
Why your AgentOps.NET project specifically? +
It's my public credibility check. Anyone can claim Microsoft Agent Framework expertise. AgentOps.NET shows specifically what I've built, the architectural decisions I've made, and the OWASP-aligned defaults I ship by default. If you read the README and the threat-model docs and they make sense for your engagement, we're aligned.
Let's Talk
Ready to ship production
agentic AI on .NET?

30 minutes. No slides, no pitch deck — just a technical conversation about your MAF / MCP / Foundry production challenge. You'll leave with concrete next steps and an honest read on whether this engagement is the right fit.

Book a 30-Minute Strategy Call →
No obligation. If your problem isn't a fit for senior .NET + agentic AI specialist work, I'll tell you straight.
Book a Strategy Call →