Production retrieval-augmented-generation platform on Terraform-provisioned Azure. A multi-stage pipeline — dense vector search over Azure AI Search followed by cross-encoder reranking — lifted retrieval accuracy ~40% over a naive top-k baseline, served through FastAPI microservices with GDPR-compliant data handling.
Awais Ahmed
AI Engineer — from prototype to production.
Ingolstadt, Germany
AI Engineer with 3+ years of experience taking Generative AI systems from prototype to production — enterprise RAG platforms (~40% retrieval-accuracy gains), LLM fine-tuning (QLoRA, 98% structured-output accuracy), and scalable FastAPI services on Terraform-managed Azure/GCP infrastructure. Combines deep technical ownership with client-facing delivery: pre-sales PoCs, demos, and solution architecture for enterprise customers. Google-certified Professional Data Engineer, currently pursuing an M.Sc. in Artificial Intelligence; experienced in GDPR- and EU AI Act-compliant deployments.
Experience & Education
2026 – Present
N![NFON AG]()
AI Engineer · NFON AG
Building LLM-powered conversational-intelligence features — call-transcription analysis, summarization, and real-time agent assist — for a European cloud-communications platform. I design RAG and agentic workflows over product and customer-interaction data with GDPR- and EU AI Act-compliant handling, and ship the Python/FastAPI inference microservices behind them with containerized CI/CD and automated quality evaluation.
2026 – Present
T![TH Ingolstadt]()
M.Sc., Artificial Intelligence · TH Ingolstadt
Graduate study in AI, pursued part-time while working as an AI Engineer.
2023 – 2026
S![SII Technologies]()
Data & AI Engineer · SII Technologies
Architected and deployed an enterprise RAG platform on Terraform-provisioned Azure (FastAPI, Azure AI Search, Docker), lifting retrieval accuracy ~40% with a multi-stage pipeline of dense vector search plus cross-encoder reranking; fine-tuned Llama-3-8B (QLoRA/Unsloth) to 98% valid-JSON extraction and quantized to GGUF for cost-efficient self-hosted inference. I also built the FastAPI backend of an automotive recommendation engine (NLP query parsing, VIN detection, live data enrichment) through to a client-facing demo, delivered a Jira ticket-analytics solution (.NET + Power BI) and Tableau HR dashboards, and drove technical pre-sales — scoping requirements, building proof-of-concepts, and presenting architectures that converted prospects into signed projects.
2020 – 2022
e![e.solutions GmbH]()
Software Integration Engineer · e.solutions GmbH
Developed Python automation and Jenkins CI pipelines for automotive software releases, cutting manual integration effort by 50% and streamlining OEM release cycles and Artifactory artifact management.
2019 – 2023
T![TH Ingolstadt]()
B.Eng., Engineering & Management · TH Ingolstadt
B.Eng. thesis · AIMotion Bavaria (Oct 2022 – Feb 2023)
Applied machine learning and clustering (K-means, Fuzzy c-means, Affinity Propagation) to large-scale unlabelled vehicle-trajectory data from the rounD dataset — recorded at three German roundabouts — to identify functional driving patterns for smart-infrastructure and automated traffic. Found that a vehicle's spatial properties strongly predict its roundabout exit, and that velocity profiles cluster tightly by exit, making velocity a reliable predictor of future behaviour.
Applied machine learning and clustering (K-means, Fuzzy c-means, Affinity Propagation) to large-scale unlabelled vehicle-trajectory data from the rounD dataset — recorded at three German roundabouts — to identify functional driving patterns for smart-infrastructure and automated traffic. Found that a vehicle's spatial properties strongly predict its roundabout exit, and that velocity profiles cluster tightly by exit, making velocity a reliable predictor of future behaviour.
Projects
A provider-agnostic sales-research agent with a custom MCP server over a mock CRM. Give it a target account and it researches firmographics, detects buying signals, finds the right decision-maker, and produces a structured brief plus a personalized outreach draft — with a code-enforced guardrail that every fact must trace back to a real tool result. Claude and OpenAI implementations swap behind one protocol.
An LLM-powered assistant that answers natural-language questions about any public GitHub repository. Built on Pydantic AI with ten focused tools over the GitHub REST API — repo overview, issues, PRs, commits, contributors, README context, and a deterministic repo-health snapshot — that the model chains automatically to answer questions like "which open issues look stale?"
Certifications
2025Google Cloud — Professional Data Engineer
2025NVIDIA Certified Associate — Generative AI & LLMs (NCA-GENL)
Writing
Notes on LLM systems, retrieval, and putting models into production.
First post in progress.