Bilal Khan

Bilal KhanJun 14, 2026

𝐅𝐫𝐨𝐦 𝐀𝐩𝐩𝐨𝐢𝐧𝐭𝐦𝐞𝐧𝐭 𝐂𝐡𝐚𝐨𝐬 𝐭𝐨 𝐀𝐮𝐭𝐨𝐧𝐨𝐦𝐨𝐮𝐬 𝐒𝐜𝐡𝐞𝐝𝐮𝐥𝐢𝐧𝐠: 𝐇𝐨𝐰 𝐈 𝐁𝐮𝐢𝐥𝐭 𝐌𝐞𝐝𝐕𝐨𝐢𝐜𝐞 𝐒𝐜𝐡𝐞𝐝𝐮𝐥𝐞𝐫.Here's the Full Technical Stack • Patients calling repeatedly about appointment times • Missed appointments costing clinics thousands in revenue loss • Language barriers delaying care • Operating hours limiting access • Manual processes bottlenecking everything One clinic receptionist told me: "I spend 60% of my day just answering scheduling calls. Our patients wait on hold forever. Worst part? Same questions, over and over." That's when I built MedVoice Scheduler — an autonomous AI voice agent that handles healthcare scheduling end-to-end. What does it do? Patients (or clinics) just call or text. The system: ✓ Recognizes the patient instantly (voice or chat) ✓ Understands their intent — book, reschedule, or cancel ✓ Checks real-time doctor availability ✓ Books, reschedules, or cancels appointments autonomously ✓ Provides instant confirmation via voice or text ✓ Operates 24/7 — no more "we're closed" In 𝑨𝒓𝒂𝒃𝒊𝒄 𝒂𝒏𝒅 𝑬𝒏𝒈𝒍𝒊𝒔𝒉. Simultaneously. How it's built 🔹 Orchestration: LangGraph workflow (extraction → verification → availability → booking) 🔹 Voice: Gemini Live + Gemini 2.5 Flash (transcription & synthesis) 🔹 Backend: FastAPI, SQLAlchemy, PostgreSQL + pgvector 🔹 Tools: MCP server (verify_patient, check_availability, book_appointment) 🔹 Multilingual: Arabic/English NLP + language detection 🔹 Frontend: React, Vite, Vapi.ai + Gemini Live voice 🔹 Infrastructure: Docker Compose, PostgreSQL pgvector, scalable architecture Why this matters Hospitals and clinics don't need more AI demos. They need AI that immediately reduces costs and improves patient outcomes. The business case is immediate: • Reduced receptionist workload • 24/7 scheduling availability • Better patient experience • Arabic + English accessibility • Reduced operational overhead For Saudi healthcare specifically: • Native Arabic voice agent • Bilingual capability • Saudi phone number handling built-in • Culturally appropriate scheduling This isn't just scheduling software. It's a template for building production-grade autonomous AI agents in healthcare. - 𝑮𝒊𝒕𝑯𝒖𝒃 repository in the comments. If you found value in the code, technical decisions, or the agentic architecture, consider giving the repository a ⭐. Healthcare needs more builders willing to solve real problems with real systems. Let's build more of this. #SeniorAIEngineer #AppliedAI #LLMSystems #VoiceAI #HealthcareTech #LangGraph #Agentic #AIEngineering #SaudiTechEcosystem Cloud Solutions

Bilal KhanJul 27, 2026

I'm pleased to share that I've successfully completed the Complete Generative AI Course with LangChain and Hugging Face Krish Naik. This course was a great opportunity to revisit and strengthen the fundamentals of Generative AI while reinforcing concepts through hands-on implementation. Throughout the course, I worked with: 🔹 LangChain for building LLM-powered applications 🔹 Hugging Face Transformers and open-source language models 🔹 Retrieval-Augmented Generation (RAG) pipelines 🔹 Prompt engineering and LLM workflows 🔹 AI application deployment strategies 🔹 Fine-tuning and integrating pre-trained models 🔹 Real-world projects, including AI assistants and intelligent applications For me, this wasn't just about earning another certificate—it was about reinforcing the core concepts that are essential for building reliable, production-ready AI solutions. As I continue my journey in AI Engineering, I'm focused on expanding my expertise in LLMs, RAG, AI Agents, cloud-native AI applications, and enterprise AI systems while continuously learning and building. A sincere thank you to Krish Naik for creating such a practical and well-structured course and for making advanced AI concepts accessible to the community. Looking forward to applying these learnings to more real-world projects. #GenerativeAI #AI #LLMs #LangChain #HuggingFace #RAG #ArtificialIntelligence #MachineLearning #AIEngineer #Python #MLOps #AzureAI #OpenSource #ContinuousLearning

Bilal KhanJun 1, 2026

𝐂𝐥𝐚𝐮𝐝𝐞 𝐎𝐩𝐮𝐬 𝟒.𝟖 𝐒𝐞𝐭𝐮𝐩 𝐆𝐮𝐢𝐝𝐞 The first thing to understand about Claude Opus 4.8 is Simple. You do not get the best result by only selecting the new model. You get the best result when you set it up properly. Model. Effort. Context. Tools. Permissions. Workflow. Cost control. That is the real setup. Opus 4.8 is not a dramatic new product that asks you to learn everything again. It is more like a sharper version of Opus 4.7 with better judgment, better long-task behavior, better tool use, and more control over how much effort Claude spends on each task. That control matters. Because one task may need a quick answer. Another task may need deep thinking. Another task may need Claude Code to inspect a whole project, run tests, find issues, and come back with something you can actually use. This guide is about setting it up for that kind of work. Not just trying it once. Actually using it. Start with the simple model choice For normal Claude users, the setup is easy. Open Claude. Pick Claude Opus 4.8. Use it for the work where quality matters. That means: complex writing |deep research | coding | planning | document analysis | business strategy | long workflows |agentic tasks | multi-step decisions Do not use Opus 4.8 for every tiny thing. That is the first cost mistake. For a quick rewrite, short email, simple summary, or small formatting task, Sonnet or Haiku may be enough. Opus is for the work where a better answer actually saves time, money, or mistakes For developers, the model ID is: claude-opus-4-8 That is the one you use in the API. #ClaudeAI #Anthropic #GenerativeAI #AIEngineering #MachineLearning #LargeLanguageModels #AIAgents #PromptEngineering #AITools #SoftwareDevelopment #AIWorkflow #DevTools #BuildInPublic #TechCommunity #AIProductivity #LLM