[ AI AGENT & FULL-STACK ENGINEER ]
ADITYA
CHAUDHARY
M.Sc. Artificial Intelligence at BTU Cottbus. I build and ship AI agent systems end-to-end — from LangGraph multi-agent workflows and MCP tool pipelines to production RAG and document-ingestion services.
> turning real problems into shipped software_
BERLIN, DE
--:--:--
↓ SCROLL
LANGGRAPH ✦ LANGCHAIN ✦ CLAUDE AGENT SDK ✦ MCP ✦ RAG PIPELINES ✦ FASTAPI ✦ NEXT.JS ✦ POSTGRESQL ✦ PYTORCH ✦ DOCKER ✦
LANGGRAPH ✦ LANGCHAIN ✦ CLAUDE AGENT SDK ✦ MCP ✦ RAG PIPELINES ✦ FASTAPI ✦ NEXT.JS ✦ POSTGRESQL ✦ PYTORCH ✦ DOCKER ✦
01 // PROFILE
I ship AI agent systems with a high-agency, ship-first mentality — most recently winning the 2026 Build with AI Hackathon with a full-stack hospital platform running 8+ live LLM workflows.
My focus is making agents smarter, cheaper, and more reliable — automating manual ops work with production RAG, document-ingestion pipelines, and multi-agent orchestration backed by FastAPI / Next.js services and Dockerised deployments.
LOCATIONBerlin, DE
FOCUSAI Agents · GenAI
DEGREEM.Sc. AI
LANGUAGESEN fluent · DE B1
AWARD '26Hackathon Winner
02 // EXPERIENCE
May 2026 — Present
Berlin, Germany
Hybrid
Hybrid
Full-Stack GenAI & Platform Engineer · Intern
Belzir GmbH
- Shipped EmailGuard live — a production multi-tenant SaaS (FastAPI + PostgreSQL + React/Vite/MUI) with an admin dashboard for tenant & user management, Keycloak SSO, and branded invite + custom password-reset flows via Microsoft Graph.
- Built AI-assisted SOC tooling for the in-house SIEM: a human-in-the-loop auto-close workflow and smart triage gate on Claude Haiku that routes ~60–70% of requests to a local fallback engine — cutting LLM spend to ~$0.18/day at ~90% cache hit.
- Led a production-readiness security audit across 4 backend services — resolving every critical & high finding: database auth + network isolation, JWT rotation, a leaked client secret, and CORS / Helmet / RBAC hardening.
- Run a 7-VM private cloud (VMware vSphere): Keycloak IAM federated with Microsoft Entra ID, an OpenXPKI PKI with ACME automation, and Zabbix + Grafana monitoring — owning infra decisions end-to-end.
FastAPI
PostgreSQL
React / TypeScript
Claude Haiku
Keycloak · IAM
OpenXPKI · Zabbix
Docker · vSphere
Mar 2024 — Sep 2024
Remote
Machine Learning Intern
Pixlia Tech
- Built and deployed LLM-powered REST APIs with FastAPI — streaming responses, structured output parsing, and robust error recovery.
- Designed end-to-end LLM & ML pipelines (Python, PyTorch, scikit-learn); delivered benchmark reports that informed model-selection and prompt-design decisions.
- Developed modular, reusable AI components and prompt templates; translated technical findings into clear reports for non-technical stakeholders.
Python
PyTorch
scikit-learn
FastAPI
03 // STACK
// AGENTS
// GENAI
// LANGUAGES
// BACKEND
// FRONTEND
// INFRA & DATA
04 // EDUCATION
Apr 2023 — Present
M.Sc. Artificial Intelligence
BTU Cottbus, Germany
Machine Learning · AI Systems · Distributed Systems · Data Analysis · Modelling of Perception & Action
Apr 2018 — Jun 2022
B.Tech Computer Science
Amity University, India
Data Structures & Algorithms · Software Engineering · Machine Learning
05 // PROJECTS
★ HACKATHON WINNER '26
CliniFlow
AI-Powered Hospital Management Platform
- Engineered 8+ production LLM workflows (live clinical scribe, structured extraction, imaging analysis with confidence-scored differentials, discharge-letter generation, compliance peer review).
- Dual-provider abstraction (Claude Sonnet 4.6 + Gemini 2.5 Flash) hot-swappable at runtime.
- PDF & document ingestion: pdf-parse → LLM structuring → Prisma + PostgreSQL persistence.
Next.js 14
Claude
Gemini
Prisma
Zod
VIEW REPOSITORY ↗
AGENTIC RAG
Second Brain Agent
Agentic RAG & Knowledge Assistant
- Shipped a multi-tool LangChain agent with document ingestion, vector semantic retrieval, and structured output parsing — deployed as a FastAPI service.
- Designed the retrieval-augmented pipeline end-to-end: chunking, embedding, similarity search, and context injection into LLM calls.
- A personal knowledge base that reasons over ingested content.
LangChain
FastAPI
Vector DB
Docker
VIEW REPOSITORY ↗
★ 1ST PLACE · HACKING 4 COMMUNITY
Mooody
Mental Wellness Companion App
- Built an Android app bridging the gap between mental health professionals and patients — addressing the fact that most people experience stress and depression but avoid seeking professional help.
- Mood Booster recommends films and music based on the user's current emotional state; Self Checkup guides reflection across sleep, nutrition, exercise, and social connection.
- Meditation simulator with ambient soundscapes, plus curated mental-health articles and infographics.
Flutter
Firebase
News API
Adobe XD
VIEW REPOSITORY ↗
MediPal
AI Health Assistant & Symptom Checker
- Flutter health assistant combining AI chat, symptom assessment, medication management, and wearable insights, built privacy-first.
- 24/7 context-aware AI chat (DeepSeek API) with age-appropriate safety modes and emergency support.
- Health Connect integration for steps, heart rate, sleep, and activity, surfaced through a customizable dashboard.
Flutter
Dart
DeepSeek API
Health Connect
Provider
VIEW REPOSITORY ↗
RESEARCH · COMPUTER VISION
Driver FER
Video-Based Facial Emotion Recognition for ADAS
- Extended the lightweight DALDL CNN with a temporal-aggregation module (1D conv over grouped NIR frames) to model expression transitions across video — not just single frames.
- Trained on the KMU-FED near-infrared driver dataset (80/20 split · 55 epochs · Adam), classifying 7 emotions for emotion-aware driver monitoring.
- Built to stay accurate yet lightweight enough for real-time, in-vehicle ADAS inference.
Python
PyTorch
CNN · 1D Conv
KMU-FED · NIR
Computer Vision
VIEW REPOSITORY ↗
06 // CONTACT
LET'S BUILD
SOMETHING.
Open to roles automating ops work and making agents smarter, cheaper, and more reliable. Let's talk.