🛡️ AEGIS — Emergency System
Multilingual AI emergency guidance system bridging distressed citizens and responders via Telegram, Mistral Large LLMs, and real-time dashboard triage.
OPEN LAYER 2Warsaw, Poland · 52.23° N
I build systems that
reason about risk
and study the universe
on the side.
Chetan Bavdhankar — Expert Data Scientist & AI Engineer at ING Hubs Poland. From supernova research to AML transaction monitoring: I bring scientific rigor to data pipelines, agentic AI systems, and analytics at scale.
SCROLL TO REDSHIFT
01 — Signal
Overview
From supernova to software. With a Master's degree in Physics from Pune University and research experience in Astrophysics and Cosmology at the National Center for Nuclear Research, I bring a unique analytical perspective to data science and AI engineering.
Currently at ING Hubs Poland in Warsaw, I execute advanced analytics projects optimizing model thresholds, build robust data pipelines, and lead quarterly back-testing initiatives across 35+ entities.
My passion lies in creating intelligent systems — from multi-agent AI pipelines to conversational chatbots — that solve real-world problems with elegance and efficiency.
- Domain
- AML transaction monitoring, regulated AI
- Method
- Multi-agent pipelines, RAG, ML at scale
- Stack
- Python · PySpark · LangGraph · llama.cpp
- Scale
- 35+ entities · 70%+ cycle-time reductions
02 — Worldline
Experience
JAN 2024 — PRESENT
Expert Data Scientist
ING Hubs Poland · Warsaw, Poland
Drive B&U projects (Risk UAT, Trigger-Based BackTesting, Segmentation) to optimize model thresholds and prevent false positives/negatives, applying ML techniques including SVM, DBSCAN, and Random Forest on Transaction Monitoring data. Lead quarterly BackTesting execution across 35+ entities — managing the codebase, assigning dev tasks, and delivering PySpark pipelines over Parquet, Iceberg, and JSON sources, with auditable automations that cut execution time from 7 weeks to under 2.
Pilot developer for Microsoft 365 Copilot and Copilot Studio: shipped multi-agent solutions wired to internal data connectors for AML analysis, narrative drafting, and methodology lookups, and onboarded the team onto the platform. Built an internal AML methodology encyclopedia converting policy documents and SME tacit knowledge into logic-explicit references — grounding data for agents, self-service docs for analysts. Designed and led a PoC on a high-friction, SME-bound agentic workflow that cut its end-to-end review cycle by ~70% (10 weeks to <2) while preserving auditability, using local LLM tooling (llama.cpp, open-source models) where data sensitivity restricts cloud inference. Contributing to the team's migration to GCP: scoping AI use cases, defining PoCs, aligning with the new stack.
JUN 2021 — DEC 2023
Senior Data Scientist
ING Hubs Poland · Warsaw, Poland
Supported methodology to optimize Transaction Monitoring Systems. Planned, tested, and managed the codebase, ensuring resilience and reliability of executions. Supported BAU project readiness (control design, RUAT, BackTesting) and presented analytical reports to business stakeholders.
MAR 2018 — MAR 2021
Research Scholar
National Center for Nuclear Research · Warsaw, Poland
Merged datasets from multiple telescopes studying the most energetic Type Ia supernovae and their host galaxies. Created Monte-Carlo simulations generating 10,000+ unique, unbiased supernova samples from correlations in observed physical properties. Trained SVM models to compare observed and simulated data via cross-correlations and estimate missing effects.
03 — Origin
Education
M.SC. PHYSICS
Master of Science — Physics
Pune University · India
Specialization in astrophysics and cosmology. Foundation in statistical analysis, computational methods, and scientific rigor that carries through all my engineering work.
2018 — 2021 · RESEARCH
Astrophysics & Cosmology Research
National Center for Nuclear Research · Warsaw
Three years of research on Type Ia supernovae and their host galaxies — multi-telescope data fusion, Monte-Carlo simulation, and ML-based comparison of observed vs. simulated populations.
04 — Constructs
Projects
🏛️ ArxivGnosis
Intelligent research assistant that discovers, ranks, and summarizes arXiv papers using LLMs, with impact filtering and automated content workflows.
OPEN LAYER 2📊 Business Segmentation Analysis
Customer segmentation with K-Means on 100K+ records for risk assessment and AML compliance, validated via Silhouette Analysis and PCA visualization.
OPEN LAYER 2💬 Conversational AI Chatbots
Agentic RAG chatbot with vector DB memory and public web UI. Fine-tuned LLMs (LLaMA 2, Mistral) via Ollama for offline voice interaction.
OPEN LAYER 2🔐 FortressAI Suite
Secure AI conversational agents with encrypted local storage, biometrics, and device automation via Accessibility Services.
OPEN LAYER 2🌟 AstroNear
Astronomy & stargazing companion with location-based celestial body identification, built on MVVM and modern Android practices.
OPEN LAYER 2🧠 BrainTeaser — IQ Offline
Offline casual games & brain-teasers with privacy-first, fully local execution and Room database persistence.
OPEN LAYER 2🧭 Hodophile Travel Agent
Multi-origin travel optimizer finding global-minimum costs via a 4-agent optimization engine, multi-modal search, and neighborhood intelligence.
OPEN LAYER 2🎮 Interactive Games Collection
Browser-based chess and Sudoku with full rule enforcement, timers, and validation logic.
OPEN LAYER 2AEGIS — Emergency System
Multilingual AI emergency guidance system that bridges the gap between distressed citizens and responders. Telegram is the citizen-facing channel, Mistral Large LLMs drive guidance and language handling, and a real-time dashboard handles responder-side triage.
Stack: Python · Mistral AI · Flask · Telegram API
ArxivGnosis
Intelligent research assistant that discovers, ranks, and summarizes arXiv papers using LLMs. Features impact filtering and automated content creation workflows for staying on top of the literature.
Stack: Python · LangChain · LLMs · arXiv API
Business Segmentation Analysis
Customer segmentation using K-Means clustering on 100K+ records for risk assessment and AML compliance. Segment identification optimized with Silhouette Analysis; results communicated through PCA visualization.
Stack: Python · K-Means · PCA · scikit-learn · Pandas
Conversational AI Chatbots
Agentic RAG chatbot with vector DB memory, context handling, and a public web UI. Fine-tuned LLMs (LLaMA 2, Mistral) via Ollama for fully offline operation with speech-to-text and text-to-speech voice interaction.
Stack: RAG · Vector DB · Ollama · LLaMA
FortressAI Suite
Secure AI-powered conversational agents on Android with encrypted local storage, biometric authentication, and device automation through Accessibility Services.
Stack: Kotlin · Jetpack Compose · Room · Firebase
AstroNear
Astronomy and stargazing companion app with location-based celestial body identification, built on MVVM architecture with modern Android development practices.
Stack: Kotlin · MVVM · Hilt · Retrofit
BrainTeaser — IQ Offline
Collection of offline casual games and brain-teasers with privacy-first, fully local execution and Room database persistence.
Stack: Kotlin · Compose · Room · Coroutines
Hodophile Travel Agent
Multi-origin travel optimizer that finds global-minimum costs via a 4-agent optimization engine (Scout, Matchmaker, and friends), multi-modal search, and neighborhood intelligence.
Stack: Vanilla JS · Agentic AI · Optimization algorithms
Interactive Games Collection
Browser-based chess and Sudoku games with full rule enforcement, timers, and validation logic.
Stack: HTML5 · JavaScript · CSS3
05 — Instruments
Skills
Data Science & Analytics
- Python · SQL
- Pandas · NumPy
- PySpark · large-scale pipelines
- scikit-learn
AI & Machine Learning
- LangChain · LangGraph
- Ollama · local inference
- Mistral AI · OpenAI · Anthropic
- PyTorch
Android Development
- Kotlin · Jetpack Compose
- Room · Hilt
- Coroutines
- MVVM architecture
Cloud & Tools
- GCP · AWS
- Copilot · Copilot Studio
- Git · GitHub
- Excel
Research & Physics
- Monte Carlo simulation
- SVM · statistical analysis
- Astrophysics & cosmology
- Scientific writing
Soft Skills
- Ownership · proactiveness
- Detail oriented
- Agile
- Leadership
06 — Deep Field
Research
NCBJ · SUPERNOVAE Ia
Energetic Type Ia Supernovae & Host Galaxies
Multi-telescope data fusion
Merged datasets from several telescopes studying the most energetic Type Ia supernovae and their host galaxies, building a unified observational sample.
NCBJ · SIMULATION + ML
Monte-Carlo Population Synthesis
10,000+ unbiased samples · SVM comparison
Built Monte-Carlo simulations exploiting correlations in observed physical properties, then trained SVM models to cross-correlate observed and simulated data and estimate missing effects.
07 — Off Axis
Beyond Work
🎤Marathi astronomy rap
🎙️Astrophysics podcasting
🏃Running — 10K, Kraków
🌌Aurora hunting
🎾Squash
✈️Travel