Loading profile…
Loading profile…
Werkstudent Applied AI Engineer
ML Research Engineer at SEVEN PRINCIPLES GROUP building GraphRAG and multi-agent tender workflows that cut screening time by 58%.
GermanyWerkstudent Applied AI Engineer at SEVEN PRINCIPLES GROUP
Sole AI engineer automating end-to-end assessment of German public-sector tenders, replacing manual document review with an automated bid/no-bid decision pipeline. Built a GraphRAG document-intelligence pipeline and a multi-agent bid-assembly system, reducing manual screening time by 58% and increasing relevant bid throughput by 30%.
Graph RAGMulti Agent SystemsMCPMongoDBPythonfastAPIPDF ProcessingMicrosoft GraphRAG
Built citation retrieval for a production RAG product and developed an SAP BTP Solution-Architect agent. Built an LLM scraping and validation pipeline and implemented GenAI observability for FastAPI/LlamaIndex pipelines.
LlamaIndexSAP UI5SeleniumPydantic
Pursuing an MSc in Data Analytics, with a thesis on temporal video understanding in vision-language models.
Completed an MTech in Mechanical Systems Design, with a thesis on representation learning on raw sensor signals.
Completed a BTech (Hons) in Mechanical Engineering.
Co-founded a medical community startup as lead data scientist and represented the company in investor pitches. Increased user engagement by 42% with a hybrid recommender and reduced churn by 36% with an ensemble model.
BERTLDAMatrix FactorizationRandom ForestGradient BoostingSMOTESHAPA B Testing
Developed an LSTM achieving 87.7% on raw data from 128 sensor channels, matching classical models built on 561 hand-engineered features.
LSTMSensor DataRepresentation Learning
Developed Boosted Fully Connected Networks by adapting BoostNet from vision to tabular data and integrated them into the TabSurvey benchmark. Benchmarked against XGBoost, LightGBM, CatBoost, and deep-tabular baselines on five datasets.
BoostNetTabSurveyXGBoostLightGBMCatBoostOptunaPython
Extended Qwen2.5-VL-7B with a cross-frame attention block in all 32 vision-encoder layers for temporal understanding. Designed a two-stage distributed DeepSpeed pipeline and evaluated on VITATECS, TempCompass, and MVBench.
PyTorchTransformersMs SwiftDeepSpeedLoRAsSlurmLMMS EvalMulti Modal LLMsVideo LLMsRAGAgentic AI
Profile sourced from Himalayas.