DEV Community: Muhammad Talha Khan

Germany Has No america.gov. So I Built an Unofficial One.

Muhammad Talha Khan — Thu, 15 Oct 2026 08:00:00 +0000

You move to Germany. You need to register your address. You open one tab, then twelve. Three PDFs. A forum thread from 2019 that contradicts all of them.

Here's the part nobody tells you: Germans have the same problem.

The answers exist. They sit on official websites, spread across the federal government, 16 states and 400 districts and cities. Each has its own portal, its own rules and its own responsible office. The US has america.gov as one front door. Germany doesn't.

So I built Ask Germany. You ask in your own language, it answers only from official sources, links every sentence to the page it came from, and tells you which office is responsible where you live.

This is how it works, and what broke along the way.

The numbers first

46,641 official pages, split into about 161,000 searchable passages
The federal service directory for all 400 districts and cities, plus about 370 city and district websites
Answers in German, English, Turkish, Arabic and Ukrainian
About $0.003 in AI cost per answer
About $90 a month in infrastructure, all in the EU

Rule one: the AI never answers from memory

A general chatbot will happily tell you how to register your address. It will also mix up Bavaria and Berlin, cite a relocation blog and invent a deadline. For residence permits and benefits, a confident wrong answer is worse than no answer.

So the model is never asked what it knows. It only gets passages retrieved from official texts, and the pipeline enforces that:

Every statement must carry a source number that points to a retrieved passage.
Statements without a valid source are removed after generation.
If no official text is close enough to the question, the visitor gets "not covered" and a link to the right official portal. No guess.

That last one matters most. Saying "I don't know" is a feature.

Where the official knowledge comes from

The backbone is the federal service directory (Servicesuche Bund, also called PVOG). It has a public search API that describes administrative services for every district in Germany, including the responsible office with address, phone and opening hours. That's what makes answers local.

On top of that, a crawler collects federal and state portals and about 370 city and district websites. It respects robots.txt and the text-and-data-mining opt-out in German copyright law, and it stays at one request per second per site. Some state portals are JavaScript apps or sit behind bot protection. I don't get around that. Those states come from the federal directory instead.

Every crawled text goes into an S3 archive first. The archive is the source of truth: when I switched embedding models, I rebuilt the whole index from it without crawling a single page again.

German breaks normal search

German glues words together. "Wohnungsgeberbestätigung" (the landlord's confirmation you need to register an address) is one word. A user might type "Bestätigung vom Vermieter". Keyword search sees nothing in common.

So compound words get split at both index time and query time, and the parts are searched too. On the questions in my eval set that depend on compounds, recall went from 0.68 to 0.88.

The search itself is hybrid: meaning (Titan Text Embeddings v2, stored as binary vectors in pgvector) plus German full-text search in Postgres, run together and filtered to the visitor's town, district, state and federal level. On a 165-question eval set in five languages, the right official source is in the top 8 results 91% of the time.

A question's journey

The browser removes tax IDs, IBANs, ID numbers, phone numbers and email addresses before anything is sent, and shows the visitor what it removed.
The browser solves a small proof-of-work puzzle instead of a CAPTCHA, and gets a short-lived anonymous session. No account.
On the server, a second pass removes names, and a safety model (Qwen3 235B) screens the question.
DeepSeek V3.2 detects the language, extracts the place and rewrites the question as a German search query in official terms.
The place becomes an official regional key through the federal directory.
Hybrid search runs, then the relevance gate decides: answer or "not covered".
DeepSeek writes a short answer in the visitor's language from the passages only, with the official German term in brackets.
Unsourced statements are removed, and the answer streams back with sources and the responsible office.

Nothing is saved. Not the question, not the answer.

Privacy was a firm rule, and it cost me

No accounts. No cookies. Questions and answers are never stored, logged or used for training.

That rule kills easy wins. A normal app would cache answers to save money. A cache stores questions, so there is no cache. Analytics would tell me which topics people ask about. That's stored data too, so the only counters are anonymous: model, tokens and cost per hour, never text.

I think it's the right trade for this audience. Many people asking about residence permits are in a vulnerable situation. "We store nothing" is a promise they can actually check.

The stack

Website: Next.js on Vercel, functions in Frankfurt
API: Python and FastAPI on AWS ECS Fargate, Frankfurt
Database: PostgreSQL 17 with pgvector on RDS, in a private network with no internet access
AI: DeepSeek V3.2, Qwen3 235B and Titan Embeddings v2 on Amazon Bedrock, Stockholm
Crawler and jobs: Python, httpx and trafilatura, as Fargate tasks
Infrastructure: one CloudFormation template

What broke along the way

The free tier ran out. I started on free Postgres tiers. The full index came to about 480 MB against a 500 MB limit, before adding anything else. I moved everything to AWS instead of cutting coverage.

Bedrock wasn't available where I wanted it. My plan was Frankfurt. On my account, Bedrock only worked in Stockholm. Still EU, so the privacy promise held, but the plan changed.

I underestimated a re-index badly. I first estimated a self-hosted embedding re-index at about a dollar and under an hour. Measured on real CPUs, it was closer to 180 CPU-hours. Lesson: measure before you quote.

Crawlers saw a different page than users. The start page said "70 official websites" in the server-rendered HTML, a hard-coded fallback. Browsers replaced it with the live number, 495. Search engines and link previews never ran that script, so they saw 70. If you care about how AI search engines describe your site, check what your HTML says before JavaScript runs.

What's next

Guide pages for the most common questions, so search engines and AI assistants have something concrete to cite. A proper German and English URL split. And a lot of listening to people who find wrong answers.

If you live in Germany, or are about to move here, try it with a question you actually had: askgermany.live. It's independent, not a government website, and not legal advice.

And if you've built retrieval over messy government data in another country, I'd love to hear how you handled it. What would you have done differently?

I built a CLI tool that writes your daily standup from git commits

Muhammad Talha Khan — Sat, 21 Feb 2026 05:25:02 +0000

I kept wasting the first 10 minutes of every morning trying to remember
what I worked on yesterday before writing my standup.

So I built standup-cli.

It does three things:

Scans your git commits from the last 24 hours
Asks what you're working on today + any blockers
Formats a clean standup in plain, Slack, or Markdown format

Install it:
npm install -g standup-cli-tool
pip install standup-cli-tool

Then just run standup from any git repo.

This is v0.1 — small and focused. Would love feedback on what to add next.

GitHub: https://github.com/muhtalhakhan/standup-cli
ProductHunt: https://www.producthunt.com/products/standup-cli

DEV Education Track: Build Apps with Google AI Studio

Muhammad Talha Khan — Wed, 16 Jul 2025 08:31:19 +0000

This post is my submission for DEV Education Track: Build Apps with Google AI Studio.

What I Built

I built NPC Forge where a user comes in and enter 4 keywords and on the basis of those keywords a NPC is born with a lore and names.

Demo

Link to Applet

My Experience

I did not know that we could do such stuff over Google AI Studio, which was surprising. However, I have built lots of apps using Firebase Studio and other scripts too. I can say Google AI Studio is a lot more fun and easier than Google Firebase Studio!!

What should you know about MCP?

Muhammad Talha Khan — Wed, 16 Jul 2025 07:38:35 +0000

Ever wonder how your favorite game character knows exactly what item to use from its inventory? Or how a smart assistant on your phone can book a flight for you? 🤔

That's where something called MCP comes in, and it's a game-changer for Python developers and the world of Generative AI.

But what is MCP?

Imagine you have a super-smart robot that can talk and answer questions. This robot is like a Generative AI model. Now, what if you want this robot to do things in the real world, like fetching you a specific book from the library or ordering a pizza? The robot, on its own, doesn't know how to interact with the library's catalog or the pizza place's online menu.

This is where the Model Context Protocol (MCP) comes to the rescue!
Think of MCP as a universal translator or a set of instructions that lets our smart robot talk to and understand different tools and data sources.

It's like giving the robot a special key that unlocks the ability to interact with the real world. Before MCP, developers had to build a custom connection for every single tool they wanted the AI to use, which was messy and time-consuming.

So, what does this mean for Python developers and Generative AI?

For Python developers, MCP is like having a superpower. Python is a popular language for building all sorts of applications, and with MCP, developers can easily connect their AI creations to the outside world. They can build AI assistants that can:
a. Interact with your computer: Imagine an AI that can organize your files, create documents, or even help you code by directly interacting with your development environment.
b. Connect to online services: Your AI could access real-time information from websites, book appointments, or manage your online accounts.
c. Work with databases and other tools: Developers can create AI agents that can analyze data from a database, generate reports, and interact with business tools.

The official MCP Python SDK makes it straightforward for developers to build these connections.

The importance of MCP in Generative AI is huge.

Generative AI models are incredibly powerful at creating text, images, and code. However, without the ability to interact with the real world, they are like brilliant minds trapped in a box.

MCP breaks down these barriers, allowing AI to not just generate content, but to take meaningful actions. This opens up a whole new world of possibilities for creating smarter, more helpful, and more autonomous AI systems.

In short, MCP is a crucial step towards creating AI that can truly assist us in our daily lives and complex tasks. It's the bridge that connects the digital brain of AI to the real world of information and tools.

17 things you absolutely should know about Grok 4

Muhammad Talha Khan — Mon, 14 Jul 2025 01:17:35 +0000

As promised that I will make a come back and will start sharing articles and posts.

here's my take on the very promising model of the current available models:

https://Inkd.in/dxscqrEz

Give it a read, and please share your queries. Also, Follow me on Medium as well to stay updated as lots of stuff is coming up very soon and very quick!

Submitted MentorSpace to Hackathon (Bolt.new)

Muhammad Talha Khan — Tue, 01 Jul 2025 21:33:30 +0000

Inspiration
We were inspired by the lack of structured, accessible mentorship for students and early-career professionals. Many learners struggle to find the right guidance at the right time. With AI advancing rapidly, we saw an opportunity to build a smarter mentorship platform that goes beyond human limitations and offers scalable, personalized support.

What it does
MentorSpace connects mentees with mentors through a smart platform that uses AI to match individuals based on skills, interests, and goals. It includes features like personalized learning plans, goal tracking, and asynchronous communication. Our long-term vision is to minimize dependency on in-person mentorship by leveraging AI-driven systems that can guide and support students autonomously.

How we built it
MentorSpace was built using React and Tailwind CSS for the frontend, with a backend powered by Node.js and Firebase for authentication, real-time data, and secure storage. The AI-based matching system and planning tracker are being developed using machine learning models that analyze user input to deliver meaningful mentor matches and customized learning journeys.

The Entry!!

Challenges we ran into
Our biggest challenge was designing a truly effective AI-based mentoring experience. It required us to rethink traditional mentorship models and understand how AI can replicate or even improve parts of the human mentorship process. Integrating intelligent tracking tools and creating a smooth, supportive user experience also required constant iteration and feedback.

Accomplishments that we're proud of
We’re proud of how far MentorSpace has come from a simple idea to a platform that’s aiming to reinvent mentorship using AI. We've built a working prototype with core functionality, and we’ve received positive feedback from early testers. Seeing the concept evolve and gain real traction has been incredibly rewarding.

What we learned
We learned how a project can evolve and reshape your thinking as you build it. Originally, we just wanted to help students find mentors — but as we explored possibilities, we realized that AI could play a much larger role. We didn’t have a detailed plan at first, but through research and testing, we discovered how AI could enable mentorship to scale in a meaningful way.

What's next for MentorSpace
Next, we’re focused on expanding MentorSpace by integrating online meeting capabilities, refining our AI matching algorithm, and improving the AI-based learning tracker. We want to ensure each student gets the most relevant mentor and consistent guidance throughout their journey. We also aim to expand into global markets and build partnerships with schools, universities, and professional communities to reach a wider audience.

Hacktoberfest 2024: Code, Collaborate, and Conquer!

Muhammad Talha Khan — Mon, 14 Oct 2024 21:11:13 +0000

Sorry! I'm Late to the Party, but I Made It!
Hey there, code fans! Guess what? Hacktoberfest 2024 is here, and I finally joined the fun!

Are you ready to dive into some awesome coding? I've got a cool project just for you. It's waiting for your magic touch.
What's Hacktoberfest, you ask? It's like a big coding party that happens every October. People from all over the world join in to make open-source projects better.

Here's the best part: you can help too! I've set up my project so anyone can join in. Whether you're new to coding or a pro, there's something for everyone.

Want to be part of this? Great! Here's how:

Check out the repository [https://github.com/muhtalhakhan/Hacktoberfest2024]
Find an issue
Make your changes
Push them back

It's that easy! Plus, you might even get a cool t-shirt for joining in.
So, what are you waiting for? Let's make some code magic together.

Happy Hacktoberfest, everyone!

Hacktoberfest2024 #OpenSource #CodingFun

my idea for buildspace s5

Muhammad Talha Khan — Mon, 24 Jun 2024 02:59:53 +0000

would love your feedback.

buildspace

Prompt Engineering, A Science or an Art?

Muhammad Talha Khan — Wed, 20 Dec 2023 19:04:34 +0000

🚀 Dive into the captivating world where science meets art! 🎨🔬

🔍 Exploring Prompt Engineering: A Science or an Art?

📅 Save the Date: 21-12-2023
🕒 Time: 08:30 PM
🌐 Location: Microsoft Teams

Join us for a thrilling event that blurs the lines between creativity and technical prowess! From keynote speeches to interactive workshops, discover the magic in engineering promptness.

Is Sam Altman’s Hiring at Microsoft Going to Make an Unveiling Change in the Industry?

Muhammad Talha Khan — Tue, 21 Nov 2023 23:10:22 +0000

Sam Altman’s recruitment at Microsoft has stirred a buzz in the tech world, sparking curiosity about the potential transformation it might usher in. This article delves into the details surrounding this significant move, analyzing its potential impact, Altman’s background, and the implications for both Microsoft and the broader industry.

Looking for help on Mistral-7B

Muhammad Talha Khan — Tue, 31 Oct 2023 20:20:38 +0000

Hey guys,
I am shifting my code from GPT to Mistral

I am facing some problems related to embedding model and engine whereas GPT has both of them publicly available.

Any help, tip or idea shared will be beneficial for me.

I also dropped a comment in the discussion of GitHub repo for Mistral-7b.

I am trying to get the help as soon as possible for finish things up at my end.

https://github.com/OpenNMT/CTranslate2/issues/1501#issuecomment-1787965398

Looking forward to your help.. Thank you!!

I got a tree from HacktoberFest 😍👾

Muhammad Talha Khan — Sat, 14 Oct 2023 11:08:13 +0000

Guys, I am one of the first 50 Thousand of the people whose their first PR/MR merged.

I am delighted that my little contributions created an impact of winning the impossible around the globe as we know there are lots of developers who are way ahead of me.

I want my fellows and beginners to always check out this kind of festivals and any of the coding competition to help them stay ahead of others.

Connect with me!!