BACK TO SELECTED WORK
LOCAL-FIRSTCLOSED-LOOP
01 // CASE STUDY/AIR-GAPPED AIOPS & SELF-HEALING INFRASTRUCTURE

AEGIS

Closed-loop, local-first SRE platform for detecting container failures, diagnosing incidents, and executing policy-controlled remediation.

ROLELead Systems Architect & Core Developer
SYSTEMDistributed SRE / AIOps Platform
RUNTIMEFully Local / Air-Gapped
STATUSActive
PRIMARY STACKNestJS · Kafka · MongoDB · Python · Docker
INSPECT SOURCE CODE
01 // THE PROBLEM

THE PROBLEM

Modern containerized systems fail through OOM conditions, timeouts, crash loops, port collisions, memory leaks, and permission errors. Detecting the failure is only the first step — automated systems must safely diagnose, evaluate policy, remediate, and audit without human bottlenecks.

DETECT

Understand what failed at the container runtime level.

DIAGNOSE

Determine the probable failure cause via local semantic classification.

DECIDE

Determine whether automated remediation is safe under strict policy.

REMEDIATE

Perform a deterministic, controlled Docker action.

AUDIT

Preserve the complete end-to-end evidence trail in durable storage.

CONTAINER FAILURE CATEGORIES
FAILURE
├── OOM
├── TIMEOUT
├── CRASH LOOP
├── PORT COLLISION
├── MEMORY LEAK
└── PERMISSION ERROR
AUTOMATED RESPONSE
Explicit failure classes parsed from stdout/stderr tails and classified by the local AI pipeline.SCROLL HORIZONTALLY ON MOBILE ↔
02 // THE APPROACH

THE APPROACH

Aegis uses a closed-loop architecture. Rather than passively alerting human operators or granting unbounded shell execution to LLMs, it connects detection to policy-gated remediation.

DETECT

Watchman catches container death and tails stdout/stderr logs.

STREAM

Publishes raw incident payloads into decoupled Kafka KRaft topics.

DIAGNOSE

Generates 384-dim log embeddings and runs FAISS + MLP classification.

EVALUATE

Validates confidence, risk level, and blast radius against safety gates.

REMEDIATE

Executes approved enum actions directly through Dockerode API.

AUDIT

Stores full incident context, embeddings, and execution receipts in MongoDB.

CLOSED-LOOP REMEDIATION WORKFLOW
┌─────────────────────┐
│ FAILING CONTAINER │
└──────────┬──────────┘
Docker Event
┌─────────────────────┐
│ DOCKER WATCHMAN │
└──────────┬──────────┘
Kafka Event
┌─────────────────────┐
│ KAFKA KRAFT │
│ EVENT BACKBONE │
└──────────┬──────────┘
┌─────────────────────┐
│ ORCHESTRATOR │
│ NestJS │
└──────────┬──────────┘
┌─────────────────────┐
│ AI ENGINE │
│ Diagnose Incident │
└──────────┬──────────┘
┌─────────────────────┐
│ SAFETY GATE │
└───────┬───────┬─────┘
│ │
PASS FAIL
│ │
▼ ▼
REMEDIATE REVIEW
DOCKER
AUDIT
Unidirectional event flow: container death through Kafka KRaft, local AI diagnosis, safety evaluation, and Dockerode execution.SCROLL HORIZONTALLY ON MOBILE ↔
03 // SYSTEM TOPOLOGY

ARCHITECTURE & DATA FLOW

Air-gapped distributed topology running entirely within the host environment. Events flow asynchronously through Kafka KRaft into local neural inference and policy-enforced actuation.

AEGIS DISTRIBUTED SYSTEM TOPOLOGY
┌─────────────────────┐
│ DOCKER HOST │
│ │
│ Failing Container │
└──────────┬──────────┘
Docker Events
┌─────────────────────┐
│ NESTJS CONTROL │
│ PLANE │
│ │
│ Docker Watchman │
│ Orchestrator │
│ Remediation │
└──────────┬──────────┘
┌─────────────────────┐
│ KAFKA KRAFT │
│ EVENT BACKBONE │
└──────────┬──────────┘
┌─────────────────────┐
│ AI ENGINE │
│ │
│ Embedding │
│ FAISS │
│ MLP Classifier │
└──────────┬──────────┘
Diagnosis
┌─────────────────────┐
│ SAFETY GATE │
└──────────┬──────────┘
┌─────────────────────┐
│ DOCKERODE │
│ REMEDIATION │
└──────────┬──────────┘
┌─────────────────────┐
│ MONGODB │
│ AUDIT / STATE │
└─────────────────────┘
Unidirectional event flow linking Docker daemon events to local AI diagnosis, safety gates, and audited remediation.SCROLL HORIZONTALLY ON MOBILE ↔
01 // CONTROL PLANE
NestJS 11 + TypeScript

Coordinates Docker Watchman events, Kafka producers/consumers, safety policy gates, and actuator calls.

02 // EVENT BACKBONE
Kafka KRaft + KafkaJS

ZooKeeper-free distributed event log for reliable incident streaming, decoupled buffering, and replay.

03 // AI COMPUTE
Python + SentenceTransformers + FAISS + MLP

Local neural network pipeline: all-MiniLM-L6-v2 embeddings, sub-ms FAISS lookup, and PyTorch MLP classifier.

04 // STATE & AUDIT
MongoDB + Mongoose

Persistent incident store, raw log history, embedding vectors, remediation plans, and offline RL replay buffers.

05 // EXECUTION
Dockerode API

Direct Unix socket client executing enum-only container actions (restart, stop, inspect) without shell scripts.

04 // EVENT FLOW

FROM CRASH TO RECOVERY

Terminal execution trace showing the step-by-step progression of an incident from container death to audited recovery.

AEGIS CORE EVENT PIPELINE TRACE9 PHASES
[01]CONTAINER.EVENT
action=dietype=containeractor.id=c8f2a1b9e0

Docker daemon emits lifecycle death signal across local Unix socket.

[02]WATCHMAN.DETECTED
source=watchman-servicestatus=abnormal_exit

Watchman listener filters out intentional stops and flags unexpected termination.

[03]LOGS.EXTRACTED
tail=100streams=stdout,stderrencoding=utf-8

Recent stdout and stderr streams extracted into raw incident diagnostic context.

[04]INCIDENT.PUBLISHED
topic=aegis.incident.detectedbroker=kafka-kraft:9092

Asynchronous incident payload committed into partitioned Kafka event backbone.

[05]ORCHESTRATOR.DISPATCH
consumer_group=aegis-orchestratorpartition=0

NestJS orchestrator consumes event and routes diagnostic payload to local AI engine.

[06]AI.DIAGNOSIS
classifier=mlpembedding=all-MiniLM-L6-v2vector_dim=384

Python service returns classification label, confidence score, and suggested action.

[07]SAFETY.GATE
confidence_threshold=0.85risk_level=LOWrule=allow_restart

Deterministic safety policy validates diagnosis against blast radius rules.

[08]DOCKERODE.ACTUATION
action=RESTART_CONTAINERdriver=docker-engine-api

Orchestrator invokes scoped Dockerode restart call directly on target container.

[09]AUDIT.PERSISTED
collection=incidentspersisted=trueoutbox_status=acknowledged

Complete audit record, diagnosis vector, and actuation receipt stored in MongoDB.

05 // SYSTEM DISCIPLINE

TECHNICAL RIGOR

Concrete systems guarantees enforced through decoupled streaming, immutable persistence, and deterministic safety boundaries.

DISCIPLINE
ENGINEERING MECHANISM
SYSTEM GUARANTEE
EVENT DELIVERY
Kafka Event Backbone
Decouples container watcher death telemetry from downstream AI diagnostic processing.
STATE
MongoDB Durable Audit Trail
Stores immutable incident records, vector embeddings, safety decisions, and receipts.
FAILURE HANDLING
Kafka Retry / Outbox Strategy
Guarantees reliable message delivery even during transient database or service hiccups.
AI
Local Inference Pipeline
SentenceTransformers (all-MiniLM-L6-v2), FAISS, and PyTorch run 100% locally without cloud dependencies.
SAFETY
Deterministic Remediation Policy
Enforces confidence thresholds (>= 0.85) and LOW risk validation before approving any actuation.
SECURITY
No Shell Execution
Eliminates command injection risks by restricting actions to Dockerode API socket calls.
NETWORK
Private Docker Bridge
Services communicate across an isolated host bridge without external egress or cloud telemetry leaks.
TESTING
Chaos Simulation Engine
Built-in failure harness injecting OOM, timeout, crash loop, permission, and port collisions.
06 // INFERENCE ARCHITECTURE

AI DIAGNOSIS PIPELINE

Aegis avoids non-deterministic generative LLMs in favor of a fast, local neural pipeline combining semantic embedding, vector similarity, and multi-layer perceptron (MLP) classification.

LOCAL VECTOR EMBEDDING & CLASSIFICATION PIPELINE
RAW CRASH LOG
PREPROCESSING
ALL-MINILM-L6-V2
384-DIM EMBEDDING
├─────────────────────┐
▼ ▼
FAISS MLP
Similarity Search Classification
│ │
└──────────┬──────────┘
DIAGNOSIS RESPONSE
Crash logs pass into local all-MiniLM-L6-v2 embedding, querying FAISS similarity and PyTorch MLP classification in parallel.SCROLL HORIZONTALLY ON MOBILE ↔
EMBEDDINGall-MiniLM-L6-v2

Preprocesses raw container stderr/stdout logs and converts them into normalized 384-dimensional dense semantic vectors.

FAISS SEARCHFacebook AI Similarity Search

Executes sub-millisecond local vector distance search over historical incident embeddings to find nearest prior failures.

MLP CLASSIFIERPyTorch Neural Network

Classifies the 384-dim log embedding into supported failure categories (OOM, Timeout, Crash Loop, Port, Perm, Leak).

DIAGNOSIS OUTPUTStructured Schema

Emits failure classification, calibrated confidence score, blast radius risk level, and suggested enum remediation action.

ENGINEERING NOTE:The entire diagnostic pipeline runs locally on CPU/GPU without cloud API dependencies, ensuring zero external data leakage and predictable millisecond response latencies.
07 // HARD PROBLEMS

ENGINEERING CHALLENGES

Key architectural challenges encountered when designing a self-healing SRE platform operating in sensitive, air-gapped container environments.

// 01SECURITY

AUTOMATED REMEDIATION WITHOUT RCE

Giving an autonomous system unrestricted shell access creates an unacceptable vulnerability if arbitrary commands enter the execution path.

// 02ACCURACY

AI UNCERTAINTY & FALSE DIAGNOSIS

A neural diagnosis can be wrong or low-confidence. Automated remediation on incorrect assumptions can amplify cluster instability.

// 03CONCURRENCY

EVENT DECOUPLING & LATENCY SKEW

Docker container death events occur in microsecond bursts, whereas local model inference has completely different latency characteristics.

// 04COMPLIANCE

ABSOLUTE AUDITABILITY

Every automated remediation decision, raw diagnostic log tail, vector similarity score, and Docker call must be fully traceable post-mortem.

// 05ISOLATION

AIR-GAPPED OPERATION

The entire platform must function without outbound internet access, cloud AI endpoints, or external model hosting infrastructure.

// 06RESILIENCE

INDEPENDENT FAILURE RECOVERY

Kafka brokers, consumer orchestrators, target microservices, or MongoDB storage can fail independently without dropping inflight incidents.

08 // ARCHITECTURAL RESOLUTION

ENGINEERING CHALLENGES & SOLUTIONS

Step-by-step engineering reasoning: Problem → Constraint → Engineering Response → Result applied across every critical reliability boundary.

// 01

AI-DRIVEN REMEDIATION WITHOUT RCE

PROBLEM

AI model output should never become arbitrary shell execution on the host container environment.

CONSTRAINT

Automated actions must remain strictly deterministic, scoped, and bounded.

ENGINEERING RESPONSE

Used an enum-only action registry (RESTART_CONTAINER, STOP_CONTAINER, IGNORE) mapped directly to explicit Dockerode Unix socket calls.

RESULT

Zero AI-generated shell scripts enter the execution path. Remote code execution is architecturally impossible.

AI-DRIVEN REMEDIATION WITHOUT RCE // ARCHITECTURE
AI DIAGNOSIS
ENUM-ONLY ACTION REGISTRY
├── RESTART_CONTAINER ──► dockerode.getContainer().restart()
├── STOP_CONTAINER ──► dockerode.getContainer().stop()
└── IGNORE ──► no_op()
Formal resolution architecture implemented inside the Aegis runtime.SCROLL HORIZONTALLY ON MOBILE ↔
// 02

LOW-CONFIDENCE DIAGNOSIS & FLAPPING

PROBLEM

Inaccurate or low-confidence diagnosis can trigger unsafe automation loops and container thrashing.

CONSTRAINT

Remediation should execute autonomously only when classification certainty is mathematically high.

ENGINEERING RESPONSE

Enforced strict threshold gates: requires confidence >= 0.85 AND risk == LOW AND action == RESTART_CONTAINER. Otherwise, automated action is skipped and flagged for operator review.

RESULT

Unsafe or uncertain actions fail-safe into human oversight rather than causing cascading container failures.

LOW-CONFIDENCE DIAGNOSIS & FLAPPING // ARCHITECTURE
EVALUATE DIAGNOSIS
├── confidence >= 0.85 ?
├── risk == LOW ?
└── action == RESTART_CONTAINER ?
┌───┴───┐
YES NO
│ │
▼ ▼
EXECUTE SKIP + OPERATOR REVIEW
Formal resolution architecture implemented inside the Aegis runtime.SCROLL HORIZONTALLY ON MOBILE ↔
// 03

EVENT DECOUPLING & BURST RESILIENCE

PROBLEM

Docker event capture should not block on AI inference or drop events during container failure storms.

CONSTRAINT

Watcher latency must remain sub-millisecond even when neural networks take longer to compute.

ENGINEERING RESPONSE

Kafka KRaft acts as the distributed event backbone between the Docker Watchman and the NestJS orchestrator.

RESULT

High-throughput event ingestion with zero dropped incident signals and decoupled consumer processing.

EVENT DECOUPLING & BURST RESILIENCE // ARCHITECTURE
DOCKER WATCHMAN (Events)
KAFKA KRAFT (Buffer Queue)
NESTJS ORCHESTRATOR (Compute)
Formal resolution architecture implemented inside the Aegis runtime.SCROLL HORIZONTALLY ON MOBILE ↔
// 04

EXPLAINABLE & AUDITABLE POST-MORTEMS

PROBLEM

Automated remediation must remain completely explainable and auditable long after execution.

CONSTRAINT

Every input, neural vector, decision gate, and API execution receipt must be immutably preserved.

ENGINEERING RESPONSE

Persisted structured incidents, raw log tails, FAISS vectors, remediation plans, and execution receipts in MongoDB.

RESULT

Complete forensic audit trail for every automated SRE event across the infrastructure.

EXPLAINABLE & AUDITABLE POST-MORTEMS // ARCHITECTURE
INCIDENT
├── RAW LOG TAIL (stdout/stderr)
├── 384-DIM EMBEDDING (FAISS)
├── SAFETY POLICY VERIFICATION
└── DOCKER ACTUATION RECEIPT
MONGODB (Persisted)
Formal resolution architecture implemented inside the Aegis runtime.SCROLL HORIZONTALLY ON MOBILE ↔
09 // ARCHITECTURAL CHOICES

TECHNICAL DECISIONS

Compact decision records detailing why each core technology was selected to guarantee air-gapped autonomy, strict isolation, and high performance.

Kafka KRaft

EVENT STREAMING
WHY

Durable event streaming and high-throughput decoupling between container event capture and neural processing.

WHY KRAFT: Operates with internal Raft quorum, eliminating Apache ZooKeeper as an operational dependency.

MongoDB

STATE & AUDIT
WHY

Flexible document persistence perfectly suited for semi-structured incident payloads, raw log arrays, high-dimensional vector embeddings, and offline RL replay buffers.

Dockerode

CONTAINER RUNTIME
WHY

Direct Docker Engine Unix socket communication without spawning shell subprocesses, eliminating shell injection and command interpretation risks.

FAISS

VECTOR SEARCH
WHY

Ultra-fast local nearest-neighbor similarity search over historical incident embeddings without cloud vector database latency or cost.

SentenceTransformers

LOCAL EMBEDDINGS
WHY

Compact, efficient local model (all-MiniLM-L6-v2) generating 384-dimensional dense semantic vectors entirely offline on CPU/GPU.

NestJS

CONTROL PLANE
WHY

Enterprise-grade TypeScript modular architecture with strict dependency injection, separating watchers, event buses, safety policies, and actuators.

10 // POLICY GATE

AUTOMATION WITH GUARDRAILS

Aegis separates diagnosis from actuation. AI recommends an action, but a deterministic policy engine evaluates safety thresholds before a single container is touched.

DETERMINISTIC SAFETY GATE DECISION TREE
AI DIAGNOSIS
┌────────────────┐
│ FALLBACK? │
└───────┬────────┘
│ NO
┌────────────────┐
│ CONFIDENCE │
│ >= 0.85 │
└───────┬────────┘
│ YES
┌────────────────┐
│ RISK == LOW │
└───────┬────────┘
│ YES
┌────────────────┐
│ ACTION == │
│ RESTART │
└───────┬────────┘
│ YES
AUTO-REMEDIATE
ANY FAILURE
SKIP
OPERATOR REVIEW
Strict hierarchical policy gates: any threshold violation or high-risk classification immediately skips actuation and alerts operators.SCROLL HORIZONTALLY ON MOBILE ↔
CONFIDENCE CEILING

Neural classification confidence must be strictly ≥ 0.85. Low-confidence ambiguity defaults to safe inaction.

BLAST RADIUS LIMIT

Risk rating must be strictly LOW. Destructive or stateful services are blocked from autonomous intervention.

FAIL-SAFE OPERATOR ROUTING

Any gate failure aborts automated execution, persists full audit telemetry, and escalates to human on-call engineers.

11 // TRUST & BOUNDARIES

SECURITY BOUNDARIES

Aegis enforces rigid execution firewalls between the probabilistic AI pipeline, the NestJS control plane, and the Docker container host.

RIGID ISOLATION BOUNDARIES
AI
│ enum only
ORCHESTRATOR
│ safety policy
DOCKERODE
│ restricted API
CONTAINER
Unidirectional privilege escalation prevention: AI emits enums only, Orchestrator checks policy, Dockerode executes API only.SCROLL HORIZONTALLY ON MOBILE ↔
NO SHELL EXECUTION

No bash, sh, or exec wrappers. Eliminates command injection vectors completely.

NO CLOUD AI

Zero outbound network calls to third-party AI APIs; fully isolated from public internet.

NO EXTERNAL LLM

Guaranteed reproducibility with deterministic embeddings and calibrated MLP output.

FIXED ACTION REGISTRY

Hardcoded enum values (RESTART, STOP, IGNORE) bounding actuator scope.

PRIVATE SERVICE NETWORK

Internal Docker bridge network with no external ports exposed for telemetry.

OPERATOR REVIEW GATES

High-risk, stateful, or uncertain incidents require explicit human sign-off.

12 // LIFECYCLE OBSERVABILITY

SYSTEM INTERACTION STATES

Deterministic incident state machine tracking every container transition from failure detection to audited recovery. Hover or click any state to inspect its operational invariant.

SAFETY_EVALUATION|normal

NestJS Orchestrator tests confidence >= 0.85, risk == LOW, and action == RESTART.

INCIDENT LIFECYCLE STATE MACHINE
HEALTHY
│ crash
DETECTED
INCIDENT_CREATED
LOGS_EXTRACTED
DIAGNOSING
DIAGNOSIS_READY
SAFETY_EVALUATION
┌──┴───────┐
▼ ▼
APPROVED SKIPPED
│ │
▼ ▼
REMEDIATED REVIEW
AUDITED
Deterministic incident lifecycle: branching at SAFETY_EVALUATION into either automated remediation or human SRE review.SCROLL HORIZONTALLY ON MOBILE ↔
13 // DATA PERSISTENCE

THE AUDIT TRAIL

Aegis is not simply a container restart script. It maintains an immutable, multi-dimensional MongoDB audit history preserving full incident context, neural vectors, and execution logs.

MONGODB ENTITY RELATIONSHIPS
SERVICE
INFRASTRUCTURE EVENT
├── INCIDENT EMBEDDING
└── REMEDIATION PLAN
ACTION EXECUTION
Relational structure linking services to raw infrastructure events, high-dimensional embeddings, plans, and execution receipts.SCROLL HORIZONTALLY ON MOBILE ↔
EPISODESOffline RL Replay Buffer

Sequential incident-state-action tuples captured to train offline reinforcement learning policies without production interference.

METRICS SNAPSHOTSInfrastructure Measurements

Point-in-time CPU, RAM, and container network metrics recorded before, during, and after remediation actions.

OUTBOX EVENTSDurable Kafka Event Retry

Transactional outbox pattern guaranteeing that every detected container incident is eventually delivered to Kafka even if brokers temporarily disconnect.

14 // ADAPTIVE POLICIES

OFFLINE LEARNING LOOP

Aegis incorporates an offline Reinforcement Learning (RL) replay pipeline to analyze historical incident episodes and evaluate future remediation heuristics without live cluster risk.

OFFLINE RL RESEARCH PIPELINE
HISTORICAL INCIDENTS
REPLAY DATASET
OFFLINE RL TRAINING
POLICY EVALUATION
RESEARCH / FUTURE POLICY
Historical incident episodes are compiled into offline replay buffers for research and future policy evaluation.SCROLL HORIZONTALLY ON MOBILE ↔
CRITICAL ARCHITECTURAL BOUNDARY: OFFLINE RESEARCH ONLY
  • The RL engine runs strictly offline in Python as a batch research tool.
  • It does NOT have direct access to the Docker socket and cannot directly restart or stop containers.
  • It can NEVER bypass or override the deterministic NestJS safety policy gate.
15 // FAULT INJECTION

CHAOS TESTING

Aegis ships with a built-in chaos engineering suite to intentionally inject container failure modes and verify closed-loop detection, neural classification, and policy gating.

CHAOS INJECTION TO REMEDIATION PIPELINE
CHAOS TRIGGER
CONTAINER FAILURE
AEGIS DETECTION
DIAGNOSIS
SAFETY GATE
REMEDIATION
AUDIT
Controlled failure injection simulating microservice crashes to continuously validate the self-healing loop.SCROLL HORIZONTALLY ON MOBILE ↔
CHAOS CLI HARNESSEXECUTABLE
$ aegis chaos oomInjects heap exhaustion
$ aegis chaos timeoutTriggers socket stall
$ aegis chaos crashForces panic termination
$ aegis chaos portInduces port collision
SUPPORTED FAILURE TAXONOMY
OOM //

Spawns a memory-greedy worker allocating unconstrained RAM until Linux kernel OOM-killer fires.

TIMEOUT //

Injects deadlocks and hung HTTP keep-alive loops that exceed service health-check deadlines.

CRASH //

Forces unhandled runtime panics and fatal SIGSEGV exit codes to test immediate recovery.

PERMISSION //

Revokes host volume filesystem permissions to evaluate permission-denied classification.

PORT COLLISION //

Attempts binding already-occupied host TCP ports to test address-in-use detection.

16 // VERIFIED ARCHITECTUREVERIFIED ENGINEERING OUTCOMES

MEASURABLE RESULTS

Grounded engineering outcomes built directly into the codebase and validated via chaos testing suites rather than speculative marketing numbers.

01

LOCAL-FIRST

Zero runtime dependency on cloud AI or external endpoints; all model inference executes on host silicon.

02

AIR-GAPPED

All services operate inside an isolated Docker bridge network with zero outbound internet egress.

03

AUDITABLE

Every raw crash log, FAISS vector distance, neural classification, and remediation action is persisted in MongoDB.

04

POLICY-GATED

Autonomous container actions strictly restricted by confidence (>= 0.85) and LOW blast radius risk thresholds.

05

DETERMINISTIC

Fixed enum-only action registry mapping directly to Dockerode calls, eliminating arbitrary command execution.

06

CHAOS-TESTABLE

Built-in failure injection harness actively verifying OOM, timeout, crash loop, and permission scenarios.

17 // COMPLETE SYSTEM TOPOLOGY

THE CLOSED LOOP

The complete autonomous lifecycle: capturing Docker failure signals, streaming through Kafka, computing local neural classifications, verifying safety gates, executing remediations, auditing state, and compiling offline learning buffers.

END-TO-END CLOSED LOOP REMEDIATION PIPELINE
FAILURE
DOCKER EVENT
EVENT CAPTURE
KAFKA
ORCHESTRATOR
AI DIAGNOSIS
SAFETY GATE
/ \
PASS FAIL
│ │
▼ ▼
REMEDIATE REVIEW
AUDIT
HISTORICAL
DATA
OFFLINE LEARNING
Unbroken closed-loop architecture connecting raw failure detection to guarded execution and offline replay datasets.SCROLL HORIZONTALLY ON MOBILE ↔
18 // ENGINEERING INTENT

WHAT I WAS SOLVING

ARCHITECTURAL CORE
“I designed a closed-loop remediation system where AI can recommend an action, but deterministic policy controls whether that action is allowed to execute.”

Autonomous systems in production cannot rely on probabilistic model outputs alone. True infrastructure safety demands decoupling recommendations from execution, bounding actuators to fixed APIs, and keeping the entire loop air-gapped.

19 // OPEN REPOSITORY

EXPLORE THE IMPLEMENTATION

Inspect the NestJS control plane, Kafka KRaft streaming consumers, Python FAISS/MLP pipelines, and Dockerode actuation modules on GitHub.

VIEW SOURCE
NEXT CASE STUDY

CarePulse

Healthcare Operations & Appointment Workflow Engine

EXPLORE NEXT