№ 00
Vol. 01
Berlin
MMXXVI
AI / ML
engineer
& consultant.
Fig. 01 — Portrait
Production NLP, recommendation systems, and analytics that actually move numbers.
"Pleasant about
deadlines, unpleasant
about magic numbers."— Mission statement, in eight words
№ 01
— A short
biography
About.
Berlin, 2026Ongoing chapter
I translate between two languages — boardroom and Jupyter notebook — without losing too much in either direction.
Data scientist with four years at Accenture's Strategy & Consulting Global Network, delivering end-to-end ML and NLP solutions for Fortune 500 clients. Advanced from Analyst to Technical Lead in three years.
The work I'm proud of lives in the unglamorous middle of the stack: BERT/RoBERTa pipelines at 90%+ accuracy on retail data, ensemble recommendation systems combining collaborative filtering with propensity scoring, market-mix models with proper adstock and saturation. I delivered a €41M revenue opportunity through NLP-based categorization and geographic analysis across 350,000+ retail outlets, and won the firm-wide GenAI Hackathon in 2024.
Now in Berlin, learning German, occasionally Spanish and French, reading papers on retrieval and representation learning.
LocationBerlin, Germany · CET'25 →
Latest roleTechnical Lead, Accenture S&C GN'21–'25
EducationMBA, IIT Delhi · CGPA 9.03'19–'21
UndergraduateB.Tech CS, DIT University'14–'18
Hackathon★ GenAI 2024 winner · Accenture'24
CertsAzure AZ-900 · Tableau · Six Sigma GB—
StatusOpen to opportunitiesnow
CurrentlyReading graph-RAG papers, learning DE'26
№ 02
— Selected
case studies
Work.
5 projects2024 – 2026
№
Project
Summary
Stack
Links
Status
01
/05
SlideKick
Strategic research copilot
A six-agent pipeline that turns a vague query into a real research deliverable — chat or full deck. Combines a Neo4j knowledge graph, web search, and financial data via MCP servers, with a quality-based refinement loop.
LangGraphLangChainNeo4jFastAPINext.jsMCP
● live
02
/05
Budget Black Hole
MMM simulator
Gamified marketing mix modeling with adstock decay, saturation curves, and a Gemini-powered analyst who critiques your channel allocation. Two modes: Sandbox & Challenge, multiple market scenarios.
PythonFlaskscikit-learnGemini APIJS
● live
03
/05
Sorting Hat
RFM & clustering tool
Full-stack customer segmentation — RFM analysis + K-Means clustering with interactive 3D visualization and auto-generated personas. Built for analysts who want the answer in under a minute.
PythonFlaskPandasPlotly.jsscikit-learn
● live
04
/05
Brand Chronoscope
Language evolution analyzer
Tracks semantic drift across 10+ years of corporate documents using sentence transformers, TF-IDF temporal trends, and dynamic topic discovery — with proper p<0.05 significance testing so changes are real, not vibes.
PythonReactFastAPISentence-BERTTransformers
● live
05
/05
Desc. Generator
End-to-end MLOps
A production-grade MLOps reference: pytest in CI, model versioning with DVC + Google Drive, BLIP vision + Gemma 2B language model, Prometheus metrics, Docker deployment on HuggingFace Spaces.
FastAPIPyTorchDVCDockerPrometheusGH Actions
● live
SlideKick01
Black Hole02
Sorting Hat03
Chronoscope04
Desc-gen05
№ 03
— Tools of
the trade
Skills.
Self-rated
Core ML 06
- Python★★★★★
- PyTorch★★★★
- scikit-learn★★★★★
- XGBoost★★★★
- pandas / NumPy★★★★★
- R★★★
NLP & GenAI 06
- BERT / RoBERTa★★★★★
- Sentence-BERT★★★★
- LangGraph★★★★
- LangChain★★★★
- RAG · Neo4j★★★★
- MCP★★★
Deploy & Data 06
- SQL (advanced)★★★★★
- FastAPI★★★★
- Docker★★★★
- Azure★★★
- Tableau★★★★
- Power BI★★★★
Languages 05
- EnglishFluent
- HindiNative
- GermanB2
- SpanishA1
- FrenchA1
