// frameworks & tooling

Forty-three frameworks. Documented.

NLP, machine learning, Azure ML & cloud, MERN + API integration, Graph RAG. Click any framework to see what it does, how I use it, and the code I ship with it.

Pipelines that turn raw, messy human language into structured signal — embeddings, intent, entities and grounded answers.

spaCy

Industrial-strength NLP pipelines — tokenization, NER, dependency parsing, custom components.

→ structured extraction from PDFs & tickets

Hugging Face Transformers

200k+ pre-trained models. Fine-tune BERT, T5, Llama, Mistral on your own corpus.

→ domain-tuned classifiers & embeddings

LangChain

Orchestrates LLMs, tools, memory and retrievers into agentic workflows that actually ship.

→ multi-step AI agents & tool use

NLTK

Classic NLP toolkit — tokenizers, stemmers, corpora, the foundation behind every linguistics course.

→ research, prototyping, education

OpenAI & Anthropic

Frontier LLMs — GPT-4 / Claude — wired with prompt caching, tool use and structured outputs.

→ production reasoning & generation

BERT & RoBERTa

Bidirectional transformers for embeddings, classification and span-level question answering.

→ semantic search, sentiment, intent

Whisper & TTS

Speech-to-text and natural voice synthesis — true to "Vaaani" (वाणी = voice).

→ voice agents, multilingual transcription

Sentence-Transformers

Sentence-level embeddings for semantic similarity, clustering and dense retrieval.

→ vector search backbone

Dify

Open-source LLM app platform — visual workflows, RAG, agent definitions, all self-hostable.

→ no-code AI workflows for clients

OpenHands

Open-source autonomous AI software developer (formerly OpenDevin) — runs in a sandboxed VM.

→ agentic coding & refactor automation

From classical regression to billion-parameter transformers — picked per problem, not per hype cycle.

PyTorch

Dynamic graphs, research-grade DL. My default for fine-tuning, distillation and custom models.

→ fine-tunes, custom transformers

TensorFlow / Keras

Production deep learning, mobile (TFLite) and edge inference for Android apps.

→ on-device models, classifiers

scikit-learn

Classical ML done right — pipelines, cross-validation, ensembles, feature engineering.

→ tabular baselines, MVPs

XGBoost & LightGBM

Gradient boosting champions for tabular data — leaderboard winners that ship to prod.

→ churn, fraud, lead scoring

Pandas + NumPy

The data backbone — vectorized math, dataframes, joins, the daily bread of every ML project.

→ ETL, EDA, feature stores

JAX

Functional, JIT-compiled, autodiff. Used when raw throughput on TPU/GPU matters.

→ research-scale training

MLflow + Weights & Biases

Experiment tracking, model registry, reproducibility. So you trust what's running in prod.

→ MLOps, model lineage

Optuna & Ray Tune

Hyperparameter search at scale — Bayesian, Hyperband, distributed.

→ wringing the last 3% of accuracy

Enterprise-grade ML, hosted on Azure (or AWS / GCP) with auth, audit, region pinning and compliance baked in.

Azure Machine Learning

Managed training, AutoML, model registry, real-time endpoints — the workhorse for enterprise builds.

→ regulated industries (BFSI, health)

Azure OpenAI Service

GPT-4 class models inside your Azure tenancy — your data stays in your region, with SLAs.

→ private LLMs for enterprises

Azure Cognitive Services

Pre-built APIs for vision, speech, language and decision — wire them up in hours, not weeks.

→ OCR, translation, form parsing

Azure Functions

Serverless inference triggers — webhooks, queues, cron — pay-per-call AI workers.

→ event-driven AI pipelines

AWS SageMaker

Train, host and monitor models with autoscaling endpoints — when the customer's stack lives on AWS.

→ multi-tenant inference at scale

Google Vertex AI

End-to-end ML on GCP — Gemini integration, BigQuery ML, AutoML Tables.

→ data-warehouse-native AI

Databricks

Lakehouse + MLflow + Unity Catalog — when data, training and governance live in one place.

→ enterprise data + AI fusion

Docker + Kubernetes

Containerized AI workers with auto-restart, health checks and zero-downtime deploys.

→ portable production deploys

The web body around the AI brain — typed APIs, secure auth, realtime updates, and a clean React UI for non-technical operators.

MongoDB

Flexible document store — perfect for nested AI artifacts (chats, embeddings, metadata).

→ chat history, RAG corpus

Express.js

Fast, minimal Node API layer — middleware pipeline, REST + JSON, easy to test.

→ public + internal APIs

React + Next.js

Component-driven UI, SSR/ISR, server actions — modern React the way it ships in 2026.

→ dashboards, marketing sites

Node.js

Async runtime, event loop, streams — perfect glue between your AI workers and your customers.

→ webhooks, queues, schedulers

REST + GraphQL

Schema-first APIs — REST for simplicity, GraphQL for typed contracts and bandwidth control.

→ public APIs, mobile clients

JWT + OAuth 2.0

Auth done properly — refresh tokens, scopes, RBAC, SSO with Google / Microsoft / GitHub.

→ multi-tenant, role-based access

WebSockets + Webhooks

Realtime chat streams, typing indicators, event push from third-party SaaS.

→ live chat, agent updates

LibreChat

Open-source ChatGPT clone built on the MERN stack — multi-model (OpenAI, Anthropic, Mistral, Ollama), self-hostable.

→ branded internal chat for customer teams

Stripe · Twilio · SendGrid

Payments, SMS/WhatsApp, transactional email — battle-tested integrations across every build.

→ checkout, OTP, alerts, drips

Beyond plain vector search — knowledge graphs + retrieval give your LLM relationships, not just nearest neighbors. The unfair advantage for complex domains.

Industries I work in

Education & EdTech Healthcare Finance & BFSI Retail Logistics SaaS Research Labs

Platforms shipped

Web (React / Next.js) Android (Kotlin / Flutter) REST + GraphQL APIs WhatsApp / Slack bots VS Code extensions