DEV Community: Tahir yamin

🛡️ Ethics Guardian: Proactive AI Governance powered by Algolia Agent Studio

Tahir yamin — Sun, 01 Feb 2026 16:00:48 +0000

This is a submission for the Algolia Agent Studio Challenge: Consumer-Facing Conversational Experiences

What I Built

Ethics Guardian is a high-stakes "Mission Control" dashboard designed for AI engineers, compliance officers, and developers. It transforms static regulatory frameworks (like GDPR, NIST AI RMF, and ISO 42001) into Live Intelligence.

Instead of searching through PDFs, developers can paste their product specifications directly into the Ethics Guard, where an Algolia-powered AI agent performs a real-time audit, identifies risks, and suggests grounded remediation steps—all in sub-50ms search time.

✨ Key Features

Advisor Relay: A conversational interface grounded in Algolia's knowledge index.
Side Panel Telemetry: Live HUD showing compliance "Safe Scores," latency metrics, and framework status.
Grounded Audits: Every suggestion links directly to official regulatory documentation (ISO, NIST, EU AI Act).
Proactive Search: Integrated Algolia Search for instant lookup of legal clauses and best practices.

Demo

🛰️ Live Interface: ethics-guardian.tahiryamin.com
📹 Video Walkthrough: Click here to watch the demo
💻 Public Repository: GitHub: algolia-ethics-guardian

How I Used Algolia Agent Studio

I implemented a Grounded Retrieval-Augmented Generation (RAG) architecture using Algolia as the central "Law Library."

1. The Grounded Agent Strategy

The core of the experience is the Advisor Relay, which uses an Algolia Agent configured with a "Search-First" system prompt.

Prompt Engineering: The system prompt is strictly engineered to prioritize toolCall outputs. It forbids the agent from answering compliance questions from its general training data. Instead, it must execute search_ethics_index to retrieve the exact ISO/GDPR Article ID before constructing a response.
Self-Correcting Loop: By treating the Agent Studio as a Model Context Protocol (MCP) server, I enabled the agent to autonomously "fact-check" itself against the latest index updates before replying.

2. Multi-Modal Integration

Dynamic "Non-Conversational" UI: In the AuditForm, I used the Algolia Agent REST API to process raw text submissions. The agent analyzes the input and returns a set of "Signal Cards" that map to high-confidence search hits in the index.
InstantSearch Integration: I utilized react-instantsearch to build the Knowledge Repo, allowing users to manually browse the frameworks that power the AI's logic.

3. Technical Implementation & MCP

I built this project on Next.js 16 (React 19 RC) to leverage the latest Server Actions for seamless AI streaming.

Google Stitch MCP: Used during development to rapidly scaffold the UI components from my design specs.
Algolia as MCP: By wrapping the Algolia API in an MCP interface, I enabled the "Mission Control" agent to autonomously call the tools it needed, effectively creating a self-driving governance engine.

Why Fast Retrieval Matters

In AI Ethics, Speed is Trust.

When a developer is auditing a feature for "Biometric Data Privacy," seeing a millisecond-fast response backed by the actual EU GDPR Article 9 text builds instant credibility. If the AI takes 10 seconds to "think," the developer loses the tactical flow of the Mission Control experience.

Algolia's lightning-fast retrieval enables our dashboard to pulse, fluctuate, and audit in real-time. It transforms "Compliance" from a bureaucratic hurdle into a live performance metric.

Challenges overcome

Hydration Mismatch: Implementing high-frequency SVG telemetry (the pulsing "Safe Score" gauge) in Next.js 16 required careful state management. I decoupled server-side pre-rendering from client-side simulation loops to ensure the dashboard remains instant and SEO-friendly.
Prompt Grounding: Crafting the logic to force the agent to cite official Framework IDs instead of general knowledge was a deep dive into Algolia's tool-use capabilities.

What's Next

Multi-Index Expansion: Adding specialized indices for global "Incident Reports" and "AI Whitepapers."
Visual Compliance Trends: Moving from a static score to a historical trend chart powered by Algolia analytics.

Acknowledgments

Built with ❤️ for the Algolia Agent Studio Challenge. Special thanks to the Algolia team for the incredibly fast search infrastructure that makes AI grounding possible.

algoliachallenge #devchallenge #ai #agents #nextjs #ethics

Aegis-OS: Industrial Intelligence HUD by Gemini 2.5-Flash

Tahir yamin — Sat, 31 Jan 2026 15:50:57 +0000

This is a submission for the New Year, New You Portfolio Challenge Presented by Google AI

About Me

I'm Tahir Yamin, an industrial Mechanical Engineer from Pakistan passionate about bringing AI to high-stakes operational environments. My background spans manufacturing automation, power plant operations, and industrial safety systems—domains where a single engineering decision can mean the difference between smooth operations and catastrophic failure.

With this portfolio, I wanted to demonstrate how Google's Gemini AI can revolutionize industrial engineering by transforming static documentation (P&ID diagrams, equipment manuals, safety protocols) into interactive, intelligent assistants available 24/7 on the factory floor.

Portfolio

🛡️ Aegis-OS: Industrial Intelligence HUD

GitHub: https://github.com/Tahir-yamin/agent-command-center

Live Interactive Demo - Set your Screen at 75% for complete view

agent-command-center-xi.vercel.app

👆 Click and interact with the Industrial HUD above - try uploading a P&ID diagram, querying the manual database, or chatting with the AI assistant!

Cloud Run Deployment (Required)

✅ Successfully Deployed to Google Cloud Run

$ gcloud run deploy aegis-os \
  --image gcr.io/qwiklabs-gcp-04-1def6ef2b7e7/aegis-os \
  --platform managed \
  --region us-east1 \
  --allow-unauthenticated \
  --port 8080 \
  --set-secrets="NEXT_PUBLIC_GEMINI_API_KEY=GEMINI_API_KEY:latest,NEXT_PUBLIC_RAG_API_KEY=RAG_API_KEY:latest" \
  --labels dev-tutorial=devnewyear2026

Deploying container to Cloud Run service [aegis-os] in project [qwiklabs-gcp-04-1def6ef2b7e7] region [us-east1]
✓ OK Deploying... Done.
  ✓ OK Creating Revision...
  ✓ OK Routing traffic...
  ✓ OK Setting IAM Policy...
Done.
Service [aegis-os] revision [aegis-os-00002-dft] has been deployed and is serving 100 percent of traffic.
Service URL: https://aegis-os-765925296978.us-east1.run.app

Deployment Components:

Multi-stage Dockerfile with Next.js standalone builds
Google Secret Manager for dual-key API management
IAM Secret Accessor role configured
Container: gcr.io/qwiklabs-gcp-04-1def6ef2b7e7/aegis-os

Note: Cloud Run deployment completed on Qwiklabs sandbox. Due to regional payment restrictions (Pakistani bank cards not accepted by Google Cloud billing), permanent hosting faces verification challenges. The Vercel embed above provides full functionality for demonstration.

What It Does

Aegis-OS leverages Gemini 2.5-Flash's multimodal capabilities to provide:

P&ID Analysis - Upload engineering blueprints for instant component identification and risk assessment
Manual RAG Search - Query thousands of pages of technical documentation in seconds
24/7 Industrial AI - Safety-trained assistant for operational support

How I Built It

Tech Stack

Frontend: Next.js 15 + TypeScript

Styling: Tailwind CSS + Custom HUD Animations

AI: Google Gemini 2.5-Flash (Vision + RAG + Chat)

Deployment: Docker + Cloud Run + Secret Manager

Architecture: Dual-key API with 5-tier retry logic

Gemini Integration

Three Multimodal Capabilities:

Vision Analysis:

const visionModel = genAI.getGenerativeModel({ model: "gemini-2.5-flash" });
const result = await visionModel.generateContent([
  { inlineData: { data: base64Image, mimeType: "image/png" } },
  { text: "Analyze this P&ID diagram..." }
]);

RAG Implementation:

const ragModel = getGenAI('rag').getGenerativeModel({ model: "gemini-2.5-flash" });
const chunks = await vectorSearch(query);
const context = chunks.join('\n');
const response = await ragModel.generateContent([
  `Context: ${context}`,
  `Query: ${query}`
]);

Industrial Persona: Fine-tuned safety-critical responses

Design Decisions

API Resilience: Built for 24/7 industrial uptime with:

Dual-key isolation (chat vs. RAG/Vision)
Smart retry with exponential backoff
Daily vs. minute-limit detection

Industrial HUD Aesthetic:

Tactical polygon borders
Biometric scanline animations
SVG noise grain overlay
RGB glitch effects

What I'm Most Proud Of

1. Production-Grade Resilience

Most demos ignore quota limits. I built a 5-tier retry system:

const isDailyLimit = errorMessage.includes("PerDay");
if (isDailyLimit) {
  throw new Error("GEMINI_DAILY_LIMIT: Quota exhausted for 24h.");
}
const delay = extractWaitTime(error) || (5000 * Math.pow(2, attempt));

2. Real-World Impact

Designed for actual plant operators:

Diagnostic time: Hours → Seconds
Safety: Instant instrumentation clarity
Expertise scaling: Junior operators get senior-level insights

3. Technical Documentation

Consolidated learnings into reusable patterns:

skills/gemini-resilience.md - API quota management
skills/industrial-hud-design.md - HUD components
.agent/workflows/gemini-quota-recovery.md - Recovery protocols

GoogleAIChallenge #GeminiAI #BuildWithAI #Industry40 #NextJS #CloudRun