Available for software engineering roles - backend systems, production reliability, cloud deployment, and AI-assisted workflows.

AI-Assisted Security Prototypeprototype

Network Guardian AI

AdGuard-connected network security prototype that polls DNS traffic, runs local heuristics before AI escalation, stores tenant-aware analysis history, and exposes the results through a FastAPI + React dashboard.

Case file

Main outcome

Moved from passive DNS blocking toward an explainable review workflow with domain history, risk categories, summaries, and anomaly fields.

Secondary signal

Reduced unnecessary AI calls by checking cache, metadata patterns, entropy features, and Isolation Forest scores before cloud analysis.

Evidence count

5 proof points

01

Problem

AdGuard can block domains, but the raw query log does not explain what happened, why a domain looks suspicious, or how patterns change over time. I wanted a system that made DNS activity easier to review without sending every domain straight to an LLM.

02

Solution

Built a FastAPI service around an AdGuard poller, SQLite-backed domain history, tenant middleware, JWT/API-key authentication, local entropy and anomaly scoring, optional Gemini/Ollama analysis, websocket updates, and a React dashboard for manual scans, stats, usage, and threat review.

03

Role

Full-stack prototype builder · Backend/API and analysis-pipeline implementer · Security dashboard and tenant workflow designer

01

Operational impact

OUTCOME 01

Moved from passive DNS blocking toward an explainable review workflow with domain history, risk categories, summaries, and anomaly fields.

OUTCOME 02

Reduced unnecessary AI calls by checking cache, metadata patterns, entropy features, and Isolation Forest scores before cloud analysis.

OUTCOME 03

Built product-grade surfaces around the prototype: auth, tenant context, developer/API routes, usage screens, billing routes, and a 12-panel dashboard.

02

System register

Implemented surfaces across ingestion, analysis, access control, persistence, interface, and testing.

  1. 01AdGuard query-log polling with deduplication and configurable polling interval
  2. 02Five-tier analysis path: cache, metadata classifier, ML heuristics, Isolation Forest, AI escalation
  3. 03FastAPI routes for health, domain analysis, history, manual history, chat, stats, auth, tenants, alerts, and developer APIs
  4. 04JWT sessions, API keys, RBAC roles, rate limiting, CORS, and security headers documented in the auth guide
  5. 05SQLite domain persistence with tenant-aware repository access and migration files
  6. 06React/TypeScript dashboard with stats, admin, usage, pricing, tenant selection, and manual analysis components
  7. 07Docker Compose setup for Network Guardian plus AdGuard Home
  8. 08Pytest suite covering heuristics, anomaly engine, auth, websocket, repository, cache, validation, alerting, and integrations
03

Verification ledger

Repository backed
Proof 01

README documents AdGuard interception, multi-layer analysis, tenant management, billing routes, developer API, and dashboard components

Proof 02

Architecture docs describe the AdGuard → FastAPI → cache/heuristics/anomaly/AI → React dashboard data flow

Proof 03

Authentication docs cover JWT, API keys, RBAC roles, rate limiting, CORS, and production hardening checklist

Proof 04

Sample data docs show domain analysis records with risk score, category, entropy, anomaly score, AdGuard metadata, and analysis source

Proof 05

Repository includes backend routes, tenant middleware, database models, Docker Compose, React components, and Tests_AI pytest coverage

04

Operating boundaries

Constraints

Conditions for responsible operation

  • Security results are review signals, not final incident-response decisions.
  • Default credentials and demo settings must be replaced before any real deployment.
  • AdGuard, Gemini/Ollama, Google Sheets, Stripe, and tenant settings depend on environment configuration.

Limitations

What the evidence does not yet establish

  • The repo reads like an ambitious prototype and needs clearer deployment evidence before being described as a live commercial SaaS.
  • Detection quality depends on traffic samples, feature choices, threshold tuning, and validation against realistic network data.
  • Some docs mention target metrics and commercial-readiness language; public claims should stick to implemented architecture and repo evidence.
05

Next field work

  1. Next 01

    Add a short demo video showing AdGuard logs becoming dashboard analysis records.

  2. Next 02

    Add screenshots for the stats dashboard, manual analysis flow, tenant selector, and auth screens.

  3. Next 03

    Add a README section that separates implemented features from future SaaS/billing goals.

Architecture note

Architecture

docs/ARCHITECTURE.md ↗
Architecture diagram 1
Architecture diagram 2
Architecture diagram 3
Architecture diagram 4
Architecture diagram 5
Architecture diagram 6
Architecture diagram 7
Architecture diagram 8

Next field note

Let's make the next system less fragile.

Open to software engineering roles across full-stack systems, platform and reliability work, workflow automation, and applied AI.

Write to Himanshu Lade →