Architecture
Microsoft Agent Framework Reference Architecture Overview
End-to-end architecture for multi-agent systems on Microsoft's Agent Framework — orchestration, governance, and production deployment patterns.
Technical Articles
Curated articles by topic — from 105+ published posts on Multi-Agent AI, FinOps, Cloud Infrastructure, Security, and Production Systems. See featured selection or browse all posts.
Multi-Agent AI & Orchestration
Migration
Why We Migrated from Google ADK to the Microsoft Agent Framework
The philosophy, architectural clarity, and vendor lock-in concerns that drove migrating 18+ production agents in 90 days.
Security
Defence in Depth for Agentic AI
The eleven-layer envelope — no two adjacent layers share a single point of failure for the same class of attack.
Interfaces
MARA: Five Interfaces
The five interface contracts that define a multi-agent runtime — registry, dispatch, memory, governance, and observability.
Evaluation
Multi-Turn Evals from First Principles
Why single-turn benchmarks fail for multi-agent systems — evaluation design for conversation, tool use, and multi-step reasoning.
A2A
A2A Protocol: Go Client
Google's Agent-to-Agent protocol implemented in Go — discovery, task management, and inter-agent communication patterns.
Financial Systems & FinOps
Cost Optimization
57% BigQuery Cost Reduction — Tata Group
Slot reservation transitions, MERGE anti-pattern elimination, and capacity-based pricing for 57% data warehouse cost reduction.
Transaction Engine
Globe: 30K+ TPS Kubernetes Transaction Platform
Idempotent processing, PCI-aligned design, error-code orchestration with exponential backoff and DLQ resilience.
Performance
Picnic: 47% Latency Reduction
Protobuf contracts, gRPC migration, and service consolidation for a social platform serving 1M+ users.
Testing
Picnic: Test Coverage & Prometheus Observability
Automated test coverage enforcement and Prometheus-based alerting for the Picnic social platform.
Lending
P2P Lending: Double-Entry Ledger Invariants
Enforcing accounting invariants in a peer-to-peer lending ledger — balanced entries, saga rollbacks, and audit trails.
Cloud Infrastructure & Databases
Database Architecture
Cloud Spanner PK Design & Write Hotspots
Primary key strategies, write distribution, and avoiding hotspots in globally distributed Spanner deployments.
Infrastructure as Code
SOC 2 Controls as Terraform Modules
Compliance as code — SOC 2, ISO 27001, and policy automation for regulated cloud environments.
Kubernetes
Azure Service Operator: Multi-Vendor Collaboration
Cloud lifecycle automation and enterprise Kubernetes patterns from the Ericsson/AT&T/Microsoft collaboration.
Observability
Microsoft Agent Framework Observability: Traces & Metrics
OpenTelemetry integration, distributed tracing, and metrics collection for multi-agent AI workloads.
Migrations
CDC & Minimal-Downtime Spanner Migration
Change data capture pipelines, Datastream integration, and zero-downtime migration strategies.
FinOps
BigQuery Slot Reservation Transition
Moving from on-demand to capacity-based slot reservations — commitment planning, autoscaling, and cost modeling.
Spanner
Spanner Interleaving: When to Use
Parent-child co-location in Cloud Spanner — when interleaving improves read performance and when it hurts write throughput.
Open Source
Airship + OpenTelemetry: 30% Ops Reduction
Integrating OpenTelemetry into the Airship bare-metal lifecycle manager — AT&T/Ericsson collaboration on telco infrastructure.
Open Source
Gocloud: Unified Cloud API Design
Multi-cloud abstraction in Go — how Gocloud unified AWS, GCP, and Azure resource management under a single API.
Security & Compliance
Security Architecture
Consolidated Security Deep-Dive
RBAC, RLS, token exchange, envelope encryption, hash-chained audit logs, and red team exercises for a multi-agent financial assistant.
Governance
RBI FREE-AI: 26 Recommendations to File Paths
Every one of the 26 RBI FREE-AI recommendations mapped to a specific file in a working multi-agent platform.
HIPAA
HIPAA as Go Interfaces
Expressing healthcare compliance requirements as type-safe interfaces — governance enforcement, audit logging, and data protection.
Audit
Audit Log Design
Hash-chained append-only logs, tamper-evident storage, and the design decisions behind production audit trails.
AI Governance
AIGP Reference Implementation
IAPP's AI Governance Professional body of knowledge mapped to working code — from risk assessment to monitoring.
Identity
SPIFFE/SPIRE Workload Identity
Zero-trust workload identity for service meshes — SVID issuance, attestation, and mTLS without secrets.
Data & AI Specializations
RAG
GraphRAG: When Graph Beats Vector
When graph-based retrieval outperforms vector similarity — trade-offs, cost, and performance for enterprise RAG.
Voice AI
Voice AI: Kinetic Multi-Language Patterns
ElevenLabs, STT/TTS pipelines, multilingual dialog, and the Kinetic India case study.
Healthcare AI
HL7 v2 Still Matters
Why HL7 v2 remains critical in medical AI — Bodh architecture, FHIR R4 integration, and cost-aware diagnostics.
RAG
Self-RAG & CRAG: When to Retrieve
Adaptive retrieval strategies — when the model should retrieve, when it shouldn't, and how to score retrieval quality.
Multilingual
Multilingual RAG with Bhashini
Cross-lingual retrieval for Indian languages — Bhashini integration, translation-aware embeddings, and code-switching handling.
Engineering Patterns
Go
Why Go for Production Agentic AI
Stdlib over libraries, single binary over framework, fail-closed defaults. The boring-on-purpose case for Go in regulated environments.
Operations
Agentic Security in Production
Twelve months of running multi-agent AI — SLIs, incident runbooks, drift detection, continuous adversarial testing, compliance posture as code.
Case Study
18 Agents in 90 Days: Lessons Learned
What worked, what was hard, and what we'd do differently. Real numbers on agent count, timeline, bugs, and governance policies.
Go
12 Go Idioms I Changed My Mind About
From error wrapping to interface design — opinions that shifted after years of production Go in cloud-native systems.
Payments
Saga Rollback When Half Succeeded
Compensating transactions, partial failure recovery, and the patterns for distributed payment processing rollbacks.
Resilience
BCP for AI: Forced Failure Drills
Business continuity planning for agentic AI — chaos engineering, failover testing, and resilience verification in production.
Go Runtime
GOMEMLIMIT & Soft GC Pacing
Soft memory caps, GC pacing in containers, and why GOMEMLIMIT changes every Go service running with a hard memory limit.
Go
embed.FS as a Deployment Unit
Bundling config, prompts, and UI assets into a single Go binary — collapsing multi-artefact deploys with embed.FS.
Go 1.23
iter.Seq — Pull Iterators in Go 1.23
Range-over-function iterators, composable pipelines, and practical patterns for the iter.Seq interface.
Speaking
Google Cloud Next 2022: Monolith → Microservices
30-minute stage talk on why most microservices migrations fail and how to structure one that ships.