I train models and build the software around them: a speech model small enough to run on a laptop, credit and fraud models with calibrated and reproducible evaluation, and desktop tools that keep your data on your machine.

BS in Data Science and Applications, IIT Madras.

Local-first tools for Windows. No account, no upload, no telemetry. Your recordings and transcripts are plain files on your disk.

→ butterflylabs

turn-detector is a 7.9M-parameter audio-only classifier that decides whether a speaker has finished or is only pausing, from the last 8 seconds of speech. It is built for Indian Hinglish: code-switched Hindi and English in both Devanagari and romanized script, Indian-accented prosody, filler words and trailing conjunctions.

Trained with 2,157 synthetic code-switched clips, it reaches 95.1% on held-out Hinglish, a gain of 32.0 points, while holding English accuracy flat at 0.938.

Each project fixes its evaluation holdout before any model exists, and every number below comes from a single scored run on that holdout.

roleflow is an open-source job-application agent: live search, per-job resume tailoring, and a browser agent that fills real applications. You approve every submission.

Genetic Disorder Prediction is my earliest public project, and still the one people fork.

Languages
Python · Java · Rust · TypeScript · JavaScript · SQL · Bash · PowerShell

Machine learning
Supervised learning · Unsupervised learning · Classification · Regression · Clustering · Ensemble methods · Gradient boosting · LightGBM · XGBoost · CatBoost · Random forests · Decision trees · Support vector machines · k-nearest neighbours · Naive Bayes · Logistic regression · Stacking and blending · Feature engineering · Feature selection · Target encoding · Dimensionality reduction · PCA · t-SNE · UMAP · Hyperparameter optimisation · Optuna · Cross-validation · Imbalanced learning · SMOTE · Focal loss · Class weighting · Model calibration · Threshold tuning · ROC-AUC · PR-AUC · Gini · KS statistic · F1 · Confusion matrices · Bootstrap confidence intervals

Deep learning
Neural networks · Feedforward networks · Backpropagation · Convolutional neural networks · Recurrent neural networks · LSTM · GRU · Transformers · Self-attention · Encoder-decoder architectures · Autoencoders · Variational autoencoders · Generative adversarial networks · Diffusion models · Residual networks · Batch and layer normalisation · Dropout · Regularisation · Transfer learning · Fine-tuning · Knowledge distillation · Quantisation · Pruning · Mixed-precision training · Learning-rate scheduling · PyTorch · TensorFlow · Keras · Hugging Face Transformers

Graph machine learning
Graph neural networks · GraphSAGE · Graph convolutional networks · Graph attention networks · Message passing · Node classification · Link prediction · Graph embeddings · Neighbourhood sampling · Community detection · Centrality measures · NetworkX · PyTorch Geometric · Deep Graph Library

Natural language processing
Text classification · Sentiment analysis · Named entity recognition · Tokenisation · Word embeddings · Word2Vec · GloVe · Sentence transformers · BERT · RoBERTa · FinBERT · DistilBERT · Topic modelling · Summarisation · Machine translation · Question answering · TF-IDF · n-gram models · Text preprocessing · Semantic similarity

Speech and audio
Automatic speech recognition · Whisper · Turn detection · Voice activity detection · Speaker diarisation · Audio classification · Keyword spotting · Text to speech · Spectrograms · Mel-frequency cepstral coefficients · Prosody modelling · Code-switched and multilingual speech · Synthetic data generation

Computer vision
Image classification · Object detection · Image segmentation · Face and gesture recognition · Feature extraction · Data augmentation · ResNet · EfficientNet · YOLO · Vision transformers · OpenCV · Pillow

Generative AI and LLM engineering
Large language models · Prompt engineering · Retrieval-augmented generation · Vector databases · pgvector · FAISS · Chroma · Pinecone · Semantic search · Embeddings · Chunking strategies · Reranking · AI agents · Multi-agent systems · Tool and function calling · Model Context Protocol · Agentic workflows · Structured output · Streaming responses · Context-window management · Token optimisation · LLM evaluation and evals · Guardrails · Hallucination mitigation · LoRA and PEFT · Instruction tuning · RLHF · Model routing · LangChain · LlamaIndex · Vercel AI SDK · OpenAI · Anthropic · Hugging Face

MLOps and deployment
Model serving · REST APIs · Flask · FastAPI · ONNX · ONNX Runtime · Experiment tracking · MLflow · Weights & Biases · Reproducible pipelines · Model versioning · Model monitoring · Drift detection · Docker · CI/CD · GitHub Actions · Batch and real-time inference · Edge and on-device inference · Inference optimisation

Data engineering and analytics
SQL · PostgreSQL · MySQL · SQLite · MongoDB · Supabase · pandas · NumPy · Polars · Dask · Apache Spark · Apache Kafka · ETL and ELT pipelines · Data modelling · Data cleaning · Data validation · Web scraping · Exploratory data analysis · Matplotlib · Seaborn · Plotly · Power BI · Excel

Statistics and experimentation
Descriptive statistics · Probability · Hypothesis testing · Confidence intervals · A/B testing · Experimental design · Bayesian inference · Regression analysis · Time-series analysis · Forecasting · ARIMA · Survival analysis · Causal inference · Monte Carlo simulation · Sampling methods

Explainability and responsible AI
SHAP · LIME · Feature importance · Partial dependence plots · Counterfactual explanations · Model interpretability · Fairness assessment · Bias detection · Model risk documentation · Basel III · IRB capital modelling · Scorecard development

Software and product
Next.js · React · Svelte · Node.js · Tauri · Tailwind CSS · REST API design · Playwright · Pytest · Vitest · Git · Linux · Windows desktop applications · Browser extensions · System design · Technical writing

LinkedIn · butterflylabs