GenAI & applied ML · 11+ years turning messy data into systems people actually use

I build AI that sits next to the people doing the work, not just another dashboard they have to interpret. My yardstick: does the output change what someone does in the next five minutes. Lately that means GenAI reasoning systems, RAG pipelines that don't fall over at cost, and local LLMs that don't need a hosted API bill.

while True:
    ship_something_useful()
    optimize_for_cost_or_performance()
    explain_it_to_a_skeptical_stakeholder()

• Reasoning systems that rank and explain, not just classify — priority engines, root-cause assistants, fault trees
• RAG + caching architectures that hold up at production volume without a runaway LLM bill
• Conversational analytics agents that turn "let me pull a report" into a 10-minute answer

GenAI & LLMs — prompt engineering, RAG, multi-agent systems, tool use / function calling, embeddings & vector search, LLM cost optimization, prompt injection defense, LangChain, LangGraph, CrewAI, Anthropic API, Azure OpenAI

Classical & deep ML — classification, regression, clustering, anomaly detection, time-series forecasting, XGBoost, LightGBM, CatBoost, CNNs, RNNs/LSTMs, transformers

NLP & vision — named entity recognition, topic modeling, machine translation, tokenization, sentiment analysis, OCR, object detection, image segmentation, OpenCV

Speech & generative media — speech-to-text, text-to-speech, Whisper, image and video generation APIs

Databricks — Unity Catalog (+ Volumes), Delta Lake, Delta Live Tables, Genie, Apps, AutoML, SQL Warehouses, serverless compute, MLflow

Data & infra — vector databases (Pinecone, FAISS), Neo4j, ETL pipelines, CI/CD for ML, OAuth, Playwright

Responsible AI & leadership — bias & fairness, AI governance, data privacy compliance, stakeholder storytelling, technical roadmap design, cross-functional leadership, conference speaking

Repos here split two ways: public ones are sanitized technique demos (RAG, local LLM inference, forecasting patterns); the rest are private architecture write-ups from real production systems, shared on request.

I write on GenAI, agentic systems, and responsible AI, and organize sessions with WiMLDS Pune on multi-agent systems.

LinkedIn · open to AI consulting engagements