Gradient Atlas
01Interactive ML course
Learn ML by building it: 24 chapters, 512 flashcards, NumPy from scratch.
Hi, I'm
ChandrimaDas
A data scientist with a background in electronics and communication engineering. I build machine learning, signal processing and LLM tools, and I ship each one as something you can open and use.
01 Selected work
Projects
Hover a project to preview it · Click to open itTap a project to open it
InDEx n/γ Discrimination
02Physics research with SINP
Neutron vs gamma-ray signals: 512-D STFT features, up to 100% accuracy.
Plugged In
03Data analytics dashboard
50,000 electric vehicles, every figure computed live from the raw CSV.
Construction Copilot
04LLM agent for CAD drawings
Ask building plans anything; 20 tools measure every number from the DXF.
Marginalia
05Retrieval-augmented Q&A
Upload a PDF; answers come only from its top 4 passages. No servers.
There's more on GitHub
Computer vision, robotics and AutoML experiments, plus hackathon work.
All repositoriesExperience
Work
-
Dec 2025 — Present
Alongside my M.Tech
Junior Data ScientistCalsoft
Clean and analyse real-world datasets end to end, and build the Power BI dashboards product teams decide with.
-
Jun 2025 — Dec 2025
AI/ML InternCalsoft
Built ML, deep learning and LLM solutions in Python, and validated models and dashboards before release.
Research
-
Mar 2025 — Jul 2026
Research
Thesis ResearcherSaha Institute of Nuclear Physics
Neutron/gamma classification for a dark-matter detector: 512-D STFT features from 200 real events, SVM from scratch, up to 100% accuracy. See the project
-
In preparation
Publication
Neutron vs gamma-ray acoustic signals, with MLSINP
Writing up the thesis results on superheated liquid detectors for publication.
Education
-
2026 — Ongoing
Alongside work
M.Tech, Robotics & Artificial IntelligenceIndian Institute of Technology Guwahati
Master of Technology in robotics and artificial intelligence, studied alongside my role at Calsoft.
-
2022 — 2026
CGPA 8.2
B.Tech, Electronics & Communication EngineeringRCC Institute of Information Technology
Bachelor of Technology. The signals and systems foundation behind my signal-processing work.
-
2020 — 2022
92%
Higher Secondary, PCMBBethune Collegiate School
Physics, chemistry, mathematics and biology.
-
2010 — 2020
93%
SecondaryBethune Collegiate School
Secondary schooling, completed in 2020.
Skills
Six areas I work across. Hover or tap a panel to open it.
Machine learning
Machine learning
Classical ML from scratch, SVMs, logistic regression, AutoML pipelines and careful model validation.
- scikit-learn
- NumPy
- TensorFlow
Computer vision
Computer vision
YOLOv8 detection, OCR, stereo depth estimation, and image and video pipelines.
- OpenCV
- YOLOv8
- EasyOCR
LLMs & Gen AI
LLMs & generative AI
RAG systems, tool calling, and LLM-assisted feature selection and error correction.
- LangChain
- Gemini
- Embeddings
Signal processing
Signal processing
STFT feature extraction, pulse localisation and physics-grade classification.
- STFT
- Spectrograms
- Matplotlib
Data & BI
Data & BI
Cleaning messy real-world data, exploratory analysis and dashboards that drive decisions.
- Pandas
- Power BI
- EDA
Shipping it
Shipping it
Interactive web dashboards, in-browser ML, Flask apps, and deployments on Netlify and Vercel.
- JavaScript
- Flask
- Netlify
- Git
Pythonscikit-learnNumPyPandasTensorFlowOpenCVYOLOv8LangChainGeminiPower BIJavaScriptFlaskMatplotlibEasyOCRNetlifyVercelGitLinuxPythonscikit-learnNumPyPandasTensorFlowOpenCVYOLOv8LangChainGeminiPower BIJavaScriptFlaskMatplotlibEasyOCRNetlifyVercelGitLinux
05 Say hello
Thanks for
tuning in.
Open to roles, research collaborations and interesting problems in machine learning, computer vision and signal processing. Email is the quickest way to reach me.
- Data science & analytics
- Machine learning
- Research collaborations
- Teaching & education
© 2026 Chandrima DasProgress is a thousand small gaps, closed.
All projects01 / 05
Interactive ML course
Gradient Atlas
Learn ML by building it, not memorising it.
24 chapters of classical machine learning as an explorable map, study route, flashcards and a Python-style console. You write each algorithm in plain NumPy, then check it against scikit-learn.
- 439linked sections
- 512flashcards
- 665quiz questions
All projects02 / 05
Physics research with SINP
InDEx n/γ Discrimination
A neutron's acoustic signature versus a gamma ray's.
My thesis work for a superheated-liquid dark-matter detector: pulse localisation, STFT features, and logistic regression and SVM written from scratch. Drop in a waveform and it classifies it in your browser.
- 512-DSTFT features
- 200real detector events
- 100%peak accuracy
All projects03 / 05
Data analytics dashboard
Plugged In
Washington State's electric vehicles, in one dashboard.
Every number, chart and written insight is calculated in the browser from the raw CSV, so a newer data export updates the whole site with no code changes.
- 50,000vehicles analysed
- 0hard-coded numbers
- 2live deployments
All projects04 / 05
LLM agent for CAD drawings
Construction Copilot
Measured, not guessed.
Ask building plans, distribution boards and circuits anything in plain English. Deterministic tools measure every number from the DXF, and hybrid search cites the exact spec clause for pass or fail checks.
- 20measurement tools
- HybridBM25 + vector search
- Offlinedesktop mode
All projects05 / 05
Retrieval-augmented Q&A
Marginalia
Ask the document, not the internet.
Upload a PDF and get answers grounded only in its most relevant passages. Parsing happens in the browser, and embeddings live in Netlify Blobs, so there's no external vector database.
- Top 4passages per answer
- 0servers to manage
- Geminiembeddings