AI ENGINEER · DATA SCIENTIST · NOIDA, IN
Production-grade AI across language, vision, and physical systems—designed to solve real problems, not just win demos.
Currently leading data science initiatives at HDFC Bank.
01 / EXPERIENCE
HDFC Bank · Noida
Developing Neev’s Agent Catalogue: a card-based interface that abstracts LiteLLM routing so teams can discover and configure autonomous agents. Built an internal Streamlit MLOps annotation platform for fine-tuning and RAG datasets.Quantiphi Analytics · Bengaluru
Executed a large-scale GCP Vertex AI Document OCR proof-of-concept, converting unstructured financial documents into structured JSON for downstream analytics.Alphadroid India · Noida
Orchestrated autonomous robot workflows with lift PLCs and GeoJSON multi-floor navigation; optimized multilingual RAG on SageMaker; moved MobileFaceNet biometrics to Flutter/TFLite at 99.5% accuracy; scaled robot updates through MQTT, ECR and Vault.IIT Guwahati · Underwater Vision
Built GroundingDINO dataset pipelines that improved underwater surveillance data quality by 35%, then reduced real-time tracking latency by 3–4 seconds using ONNX Runtime.HealthView · Jaipur
Developed NLP-driven content curation models and D3.js dashboards to improve user engagement.02 / SIDE PROJECTS & EXPERIMENTS
A collection of side projects and open-source experiments. Many more are available on GitHub ↗.
Agentic firmware that turns a $2 microcontroller into a general-purpose physical AI—capable of reasoning, sensing, and acting in the real world.
Launch project ↗An AI wellness companion combining Android Health Connect data with conversational intelligence.
Launch project ↗A multimodal assistant for extracting, indexing, and querying visual document data.
View source ↗An AI recruitment tool that parses PDFs, extracts skills with Gemini, scores matches with TF-IDF, and renders dynamic Plotly radar charts.
View source ↗03 / ABOUT
My work sits where generative models, computer vision, and hardware collide. I care about the unglamorous details—latency, reliability, interfaces, and whether people can actually use the thing.
04 / CREDENTIALS
2021 — 2023
2015 — 2019
PUBLICATION
RCRM 2024