Available for opportunities
Dhiraj Poddar
|
Building production AI platforms with multi-agent systems, hybrid RAG pipelines, and cloud-native infrastructure.
About Me
I build production AI platforms. As the first engineer at an AI startup, I designed and shipped the entire technical foundation — from Azure cloud infrastructure to LangGraph multi-agent systems to a Next.js frontend with real-time AI streaming. 4+ years across backend engineering, AI/ML systems, and cloud-native deployment.
0
Years Experience
0
Projects Shipped
0
Companies
0
Degrees
Experience
Agentic AI Engineer
MAindTec GmbH (AI Startup)
CurrentFirst Employee
Jan 2025 – PresentIngolstadt
- Core team building MAiQ AI SaaS platform: FastAPI modular monolith (DDD), LangGraph agentic workflows, Azure cloud infrastructure with Bicep IaC
- Implemented hybrid RAG pipeline — BM25 full-text + pgvector semantic search + Cohere reranking — with event-driven workers on Redis Streams, Stripe billing, and SharePoint integration
- Led and delivered a production AI agent end-to-end in 3 months for an external customer
- Built GPU-accelerated 2D technical drawing analysis with YOLO object detection, Transformer OCR, and GPT vision on AKS
- Implemented testing and evaluation pipelines for AI agents using LangSmith; set up PostHog analytics for LLM service monitoring
LangGraphFastAPIAzureNext.jsPostgreSQLRedisDocker
Working Student AI Engineer
Siemens AG
Sep 2024 – Dec 2024Munich (Remote)
- Developed and implemented various time series forecasting models and evaluated their performance
- Assessed model robustness by applying perturbation methods such as Brownian, Gaussian noise, rotation
- Implemented a dashboard application for the demonstration of model performance
PythonPyTorchTime SeriesDashboard
Master Thesis Student
Siemens AG
Mar 2024 – Aug 2024Munich
- Researched Robustness of Large Language Models
- Integrated open source NLP models from Hugging Face for machine translation, paraphrasing and tokenization
- Implemented text data augmentation methods: synonym replacement, backtranslation, paraphrasing
- Investigated evaluation metrics BERTScore, BLEURT, BARTScore on open source QA datasets
- Implemented RAG pipeline to extract information for Siemens dataset
RAGLLMsTransformersPyTorchNLTKHugging Face
Working Student Data Scientist
Siemens AG
Feb 2023 – Feb 2024Erlangen (On-site)
- Implemented and integrated AI models using TensorFlow and PyTorch, leading to 20% increase in prediction accuracy for PCB board soldering defect classification
- Maintained and deployed ML models using Siemens deployment infrastructures (AI inference server, AI monitor)
TensorFlowPyTorchComputer VisionMLOps
Software Developer
Citytech
Apr 2019 – Oct 2020Bagmati, Nepal
- Developed POS mobile applications in native Android (Java/Kotlin) used by 5+ banks
- Built dynamic form generation library from APIs, reducing form generation time by 30%
- Worked in agile development environment using Jira for project management
JavaKotlinAndroidFintechJira
Projects
CNN-Generated Image Detection
Researcher — FAU Erlangen-NürnbergUniversal detector to distinguish real vs CNN-generated images from 11 generator models
11
Generator Models
ResNet50
Primary Architecture
PyTorch
Framework
- Developed universal detector for real vs CNN-generated images across 11 different generator models
- Reproduced and extended results from S. Wang et al. paper using same models and datasets
- Used ResNet50 and GoogleNet pretrained on ImageNet for feature extraction
- Evaluated with Accuracy and Average Precision metrics; explored additional architectures and metrics beyond the original paper
PyTorchResNet50GoogleNetImageNetCNN
Dynamic Form Library
Software Developer — CitytechRuntime form generation from JSON APIs with multi-screen support
8+
Custom Field Types
2
Form Modes
30%
Dev Time Saved
- Built forms generated from JSON via API at runtime
- Supported single-screen (Normal) and multi-screen forms (ViewPager)
- Created custom layouts: TextView, EditText, Checkbox, CheckboxGroup, RadioButtons, Image, Map, Signature
- Used observer pattern with EventBus for image, fingerprint, and signature capture
JavaKotlinAndroidEventBusViewPager
MLOps End-to-End Pipeline
ML EngineerFull MLOps pipeline for gemstone price prediction with 98% accuracy
98%
Accuracy
4
ML Models
E2E
MLOps Pipeline
- Regression pipeline using LinearRegression, Ridge, Lasso, and RandomForest
- Full MLOps stack with experiment tracking, data versioning, and CI/CD
- Containerized with Docker, deployed on Azure
scikit-learnDockerMLflowAirflowDVCGitHubDagsHubAzure
Dimensionality Reduction using Deep Learning
Researcher — MANIT BhopalAutoencoder + t-SNE approach for Big Data dimensionality reduction
t-SNE
Visualization
Autoencoder
Architecture
- Addressed high-dimensionality challenges in Big Data
- Used various autoencoder models to reduce reconstruction loss
- Combined t-SNE with autoencoder to further decrease dimensionality reduction loss
Deep LearningAutoencodert-SNEPython
Wind Energy Forecasting
Researcher — MANIT BhopalTime series forecasting with ARIMA modeling in MATLAB
ARIMA
Model
MATLAB
Platform
- ARIMA modeling of long time series for wind energy prediction
- Autocorrelation and non-stationarity detection in pre-whitened time series
- Validated using magnetoencephalography recordings
MATLABARIMATime SeriesSignal Processing
Cracked Windows Image Recognition
Researcher — FAU Erlangen-NürnbergImage classification of cracked windows using CNN
0.65
F-Score
3
Image Classes
- Classification of three types of cracked window images
- Implemented using PyTorch and CNN architecture
PyTorchCNNComputer Vision
Skills
AI / Machine Learning
LangGraphLangChainLangSmithRAGpgvectorCoherePyTorchYOLOTransformersOpenCVPrompt EngineeringMulti-Agent Systems
Backend
PythonFastAPISQLAlchemyPydanticRedisCeleryGunicorn/UvicornAlembic
Cloud & DevOps
Azure Container AppsAKSAzure OpenAIKey VaultVNetDockerBicep IaCGitHub ActionsKEDAPrometheus
Frontend
TypeScriptNext.js 15React 19Redux ToolkitTailwindCSSFramer MotionStripeSSE Streaming
Databases
PostgreSQL 16Redis 7pgvectorMulti-Schema ArchitectureAlembic Migrations
Education
MSc Data Science
Friedrich Alexander Universität Erlangen-Nürnberg
2021 – 2024Erlangen, Germany
Relevant Courses
Artificial IntelligencePattern RecognitionDeep LearningML in Time SeriesExplainable AIBusiness Intelligence
BTech Computer Science
MANIT Bhopal
2014 – 2018Bhopal, India
Relevant Courses
Data Structures & AlgorithmsData MiningDigital Image ProcessingNLPOOP Design