About
I'm a data scientist who likes building things that actually ship. My work sits at the intersection of Generative AI and Agentic systems — retrieval-augmented generation, multi-agent orchestration, and Human-in-the-Loop workflows that stay reliable in production.
Currently I'm a Consultant in Data Science at Delphi Consulting Middle East (Remote), where I build healthcare chatbots, LLM evaluation pipelines with DeepEval, and HITL synthetic-data pipelines for RAG (RAKE, NLTK, CCG), Text-to-SQL, and schema-bound multi-table generation — scaled on Azure with Redis and Celery. I work with LangChain, AutoGen, CrewAI, Agno, and Hugging Face Transformers.
Before that, I was an Associate in Data Science at the same firm, where I built time series forecasting models for healthcare patient volume and revenue prediction, and developed an earlier multi-agent HITL chatbot for appointment booking and patient engagement.
Previously I was a Trainee Software Engineer at Ingram Micro India SSC, integrating Workday Studio applications over SOAP and REST and building HR Tech interfaces for Payroll and LMS. I completed an M.Tech in Artificial Intelligence at SVKM's NMIMS MPSTME (GPA 9.47) after a B.E. in Computer Engineering at SIES GST (GPA 9.18).
Based in Navi Mumbai.
Experience
Apr 2026 — Present Consultant Data Science·Delphi Consulting Middle East
Remote
- Developed a multi-agent LLM framework with Human-in-the-Loop (HITL) workflows for a healthcare chatbot, including a real-time FAQ agent with auto knowledge-base updates from SharePoint, a Postgres DB agent for SQL-backed queries, and Redis-based queuing for reliable concurrent request handling.
- Designed LLM evaluation and AI-agent testing pipelines capturing agent, team, and workflow responses; evaluated with DeepEval and custom metrics for automated validation, benchmarking, and regression testing, scaled with Redis and Celery on Azure.
- Built HITL synthetic data generation pipelines for RAG using keyword-matching-based retrieval with RAKE, NLTK, and CCG (Combinatory Categorial Grammar) for keyword and phrase extraction.
- Built HITL synthetic data generation pipelines for Text-to-SQL (TTS) via two paths: schema-only generation, and direct data-driven generation from Excel (multi-sheet and single-sheet) or a PostgreSQL DB connection, analyzing the underlying data to generate, run, and validate TTS queries.
- Built custom structured data generation pipelines that generate synthetic data conforming to a defined schema; for multi-table schemas, analyzed table relationships and enforced foreign-key (FK) binding to ensure accurate FK mapping and referential integrity across generated tables.
- Python
- LLMs
- RAG
- Text-to-SQL
- DeepEval
- PostgreSQL
- Redis
- Celery
- Azure
- SharePoint
Oct 2025 — Mar 2026 Associate Data Science·Delphi Consulting Middle East
Remote
- Built time series forecasting models to predict healthcare patient volume and extended the pipeline into a revenue-prediction model, supporting capacity planning and financial forecasting.
- Developed a multi-agent LLM framework with HITL workflows to power a healthcare chatbot for automated appointment booking and patient engagement.
- Designed and implemented LLM evaluation pipelines using DeepEval, creating custom evaluation metrics to assess response quality, relevance, and reliability.
- Python
- Time Series
- LLMs
- Multi-Agent
- HITL
- DeepEval
Jul 2023 — Jul 2024 Trainee Software Engineer·Ingram Micro India SSC Pvt. Ltd.
Mumbai, India
- Developed and integrated Workday Studio applications using SOAP and REST APIs for real-time enterprise workflows and ERP connectors.
- Built HR Tech interfaces for Payroll and Learning Management Systems (LMS) leveraging Workday Extend.
- Delivered 4+ end-to-end projects following Agile methodology, version control best practices, and stakeholder feedback cycles.
- Workday Studio
- Workday Extend
- SOAP
- REST
- ERP
Projects
Gita-LLM
A therapist bot grounded in knowledge of the Bhagavad Gita, using retrieval and language models to reason about problems and offer guidance.
- Python
- LLMs
- RAG
- Jupyter
Earning-Call-Analyzer
Analyzes earnings-call transcripts with large language models to surface themes, sentiment, and signals from financial communications.
- Python
- LLMs
- NLP
- Jupyter
NeuroNest
Personal NeuroNest experiments in Python around nested neural / agent workflows for applied AI prototypes.
- Python
- Agents
DataCronyx
AutoEDA and AutoTrainer: automated exploratory analysis and model-training workflows to shorten the path from dataset to baseline models.
- Python
- AutoEDA
- ML
- Scikit-Learn
Stress Prediction
Predicts stress from vital physiological metrics (heart rate, EDA, temperature, and related signals) collected from wearables — the applied counterpart to the Springer publication.
- Python
- Machine Learning
- Wearables
Edge Cafeteria App
Cafeteria management prototype exploring edge-oriented workflows for ordering and operations.
- Python
- Jupyter
- Edge
Publication
2026 Predictive Analysis of Stress Based on Vital Physiological Metrics
ICAIC 2025 · Lecture Notes in Networks and Systems, vol. 1830, Springer