Skip to content

whoami

Aayan Ahmed

Forward Deployed AI Engineer · Augsburg · Ingolstadt

I spent a year labelling invoices so an AI could read them. Now I build the AI.

story

Story

  1. origin new delhi Grew up here

    Rd Rajpal Public School, 2010–2020. Hindi is still the language I think in.

  2. 2020 delhi One year of B.Tech at Jamia Hamdard

    Information Technology, 2020–2021. Grade 1.5 on the German scale. Then Germany.

  3. 2022-09 ingolstadt Moved to Germany at 21. Technische Hochschule Ingolstadt

    B.Sc. Computer Science & AI, graduating 02/2027. Thesis: a scoping review of data spaces for sharing data between organisations.

  4. 2024-09 munich Data annotator at Gini

    10,000+ annotations of financial documents, training data for document-AI models. Helped refine the annotation guidelines.

  5. now augsburg Working student, AI Engineering at Teamworks AI

    Since 08/2026, after an internship 11/2025–05/2026. I build voice agents, n8n flows and on-prem deployments for construction firms.

  6. next Looking for a full-time AI / Forward Deployed Engineer role

    From 03/2027, after graduating.

work

Work

  1. View project: 01

    Bautagesbericht Pro

    A construction-site PWA with voice input for every field, in any language.Any language → formal German report
    Production · Teamworks AI[teamworks]
  2. View project: 02

    Outlook → CAPMO

    An n8n flow that turns triaged Outlook emails into construction tickets.
    Production · Teamworks AI[teamworks]
  3. View project: 03

    Employee Acquisition

    An n8n pipeline that turns unstructured German job descriptions into structured search criteria and a ranked candidate shortlist.Job description → ranked shortlist
    Production · Teamworks AI[teamworks]
  4. View project: 04

    GEO Tune-Up

    A research tool that asks four LLMs whether they cite a given URL when answering plausible user questions, and produces ranked rewrite suggestions.
    Personal R&D
  5. View project: 05

    Herbie

    An in-house HR portal for a small company with no HR. I started and led the project, and built it with one colleague on my team.
    Internal tool · Teamworks AI[teamworks]
  6. View project: 06

    On-Prem AI Platform

    An enterprise AI chat platform running entirely on a construction company's own Windows server.
    In progress · client project

bio

Bio

I was born in New Delhi and moved to Germany at twenty-one to study Computer Science and AI. Four years later I'm still here, in a flat in Ingolstadt with a fake lemon, two windows, and a habit of eating pizza brötchen at two in the morning.

I write code for a living. Python, TypeScript, SQL, an automation tool called n8n, the occasional bit of C++ when I'm bored. I taught myself most of it, and the things I build run in production. I work at Teamworks AI, an AI-forward consultancy near Augsburg that's quietly trying to drag the German construction industry — a sector that has been doing things mostly the same way since the Wirtschaftswunder — into the actual present.

My thesis is a scoping review of the claims and the evidence on data spaces: the connector tech (Eclipse Dataspace Connector, Gaia-X/IDS) that lets organisations share data without handing it over.

Before this I worked as a data annotator: more than ten thousand labels on financial documents, training data for document-AI models. I know what good training data looks like, and what it takes to get there.

I think in Hindi, write in English, and order Döner in German. My German is B2 and getting better every week; I use it at work every day.

Syed Aayan Ahmed in Berlin
me, in Berlin

I have one belief about the work and it's this: the best AI system is the one nobody talks about because it just works. Boring is a feature. It means it shipped, somebody's using it right now without thinking about it, and it hasn't broken yet. There's a particular, underrated kind of cool in that — and it's the kind I'm chasing.

lab

Lab

Four small demos of things I work with. Everything runs on this page; nothing is sent anywhere.

01 · Tokenizer comparator

Three tokenizers, one paragraph

Type or paste anything. GPT-4o, GPT-4 / 3.5 and Llama 3 each split it into tokens, and you pay per token. Text that isn't English usually splits into more pieces, so it costs more.

[ loading demo ]

02 · Who owns the data

One database, or five owners

A construction firm tracks site reports: workers, materials, weather, incidents, photos. In the first version, one platform team owns every table. In the second, each part of the business owns its own data, and they agree on a shared contract.

[ loading demo ]

03 · Schema as contract

A spoken site report becomes a German record

A worker speaks the day's report into the app. The transcript comes back messy. A strict JSON schema decides what counts as a valid record: here one field fails validation, the model tries again, and the record goes through. The same contract runs in production, turning voice notes in any language into the formal German daily report.

[ loading demo ]

04 · Embedding map

Words placed by meaning

About 115 words, each placed near words with a similar meaning. Hover over a word to light up its five closest neighbours. Where two groups meet, you can usually see why.

[ loading demo ]

contact

Contact

I'm in Augsburg, mostly writing code. I read every email I get and answer most of them. Say something.

English (C1) · Deutsch (B2)