Overview:
I am a senior ML/AI engineer and Software Engineers who build production-grade AI and
end-to-end systems. With experience in Healthcare, Law firm, and Real Estate, I specialize in
Machine Learning, Agentic AI, RAG, Document Intelligence, and large-scale MLOps on GCP,
Azure, and AWS. From initial MVP to full-scale deployment,I use LLMs (Gemini, GPT-4) and
modern data infrastructure to automate your most critical workflows.
Projects Portfolio:
Agentic AI, RAG, Chatbots & AI-Powered Analytics
I build intelligent agents that perform multi-step tasks and interpret complex data.
● Census Data Analysis Platform:
An interactive analytics platform that transforms raw Census data into interactive maps
and an immediate, AI-generated summary. Users select geographic areas (MSAs) and
Census data types (e.g., population, income), which are visualized on an interactive
map. The core feature is an AI-powered engine that generates a nuanced narrative
summary of the selected data, identifying key trends, correlations, and even data
limitations.
Tech: Python (Django), U.S. Census API, Leaflet.js, React, Gemini.
● Voice-Intel - An AI Clinical Transcriber & Co-Pilot:
A secure desktop application that records doctor-patient conversations, using a
Whisper-to-Gemini for two main critical functions:
○ Transcriber: Creates a concise, structured summary of the Doctor-patient
conversation, very helpful for immediate use in clinical documentation.
○ Co-Pilot: Analyzes the full transcript to flag potential missed follow-up questions,
patient concerns requiring deeper exploration, and key medical processes.
Tech: Electron.js, Whisper AI, Gemini LLM, Node.js, OS-level audio integration.
● AI Slack Agent:
An internal productivity assistant designed to streamline team operations and reduce
manual overhead. The agent automates onboarding workflows by guiding new
employees through customized checklists, ensuring consistency and compliance. It also
functions as a knowledge companion, instantly answering FAQs by searching across an
internal knowledge base.
By reducing repetitive tasks and enabling quick access to information, the AI Slack
Agent improves team efficiency, speeds up onboarding, and frees staff to focus on
higher-value work.
Tech: Slack API, Gemini LLM, Django backend, Vector Database for knowledge
retrieval.
● Job Sourcing Agent (Job Talk):
An end-to-end recruitment intelligence system designed to simplifiy how job postings are
analyzed and candidates are sourced. The platform ingests job descriptions in multiple
formats (PDF, DOC, etc.) and uses Gemini LLMs to generate structured JSON outputs,
including role details, salary ranges, keywords, and summaries.
From there, the system automates candidate sourcing through scraping LinkedIn and
Indeed. Profiles are deduplicated, enriched with data from Apollo and SignalHire, and
processed into a clean, standardized CSV for recruiters. A final AI-powered summary
provides users with context and insights.
Tech: Selenium, Gemini, Google ADK (agents & orchestration), Django, React, GCP.
● RAG Chatbot:
The Chatbot allows users to upload PDFs or other files, automatically processes them
into semantic embeddings, and enables natural language queries from the stored
content. The chatbot was deployed with a Django-based backend and UI, featuring PDF
upload, document selection, and conversational querying.
Tech: Gemini, LangChain (RAG + text splitting), Text Embedding model, PostgreSQL +
pgvector, Django.
● Healthcare Chatbot
The system uses LangGraph to orchestrate the main conversational workflow, ensuring
structured and consistent symptom collection. To ensure reliability, a lightweight
decision-tree fallback handles unexpected inputs or errors, maintaining uninterrupted
guidance for users. Gemini generates clear, concise responses aligned with trusted
medical guidelines, and a conversation history manager preserves context across the
chat history.
Its main features are:
● Collects symptoms through guided, structured questioning.
● Provides a decision-tree fallback to maintain reliability if the main workflow
encounters issues.
● Summarizes findings clearly for the user.
● Directs users to relevant, verified medical resources rather than generating
diagnoses.
● Maintains conversation context for coherent conversation history.
Tech: LangGraph (workflow orchestration), Decision Tree (fallback logic), Gemini (LLM),
Django (backend and web interface).
Document Intelligence & Automation:
I am experienced in extracting structured data from unstructured documents like PDFs and
images.
● AI Law Bill Processor:
A system for law firms that ingests pre-bill PDFs, automatically corrects routine errors,
and flags complex issues for human review. The AI Bill Processor is a web-based
application designed to streamline and automate the legal billing review workflow. By
combining Azure document-intelligence and large-language models (Gemini), the
platform ingests pre-bill PDF batches, automatically corrects routine errors, and
intelligently flags issues and reduced manual effort and error rates in law-firm billing
audits.
● Mortgage Legal Document Extractor:
An AI document-processing platform built to automate and accelerate real estate
closings by extracting critical information from legal, financial, and identity documents.
The system supports use cases in conveyancing, mortgage underwriting, and
compliance review by converting unstructured documents into structured, usable data.
● Building Plan OCR:
An automated OCR and Computer vision system that processes construction plan PDFs
to extract quantities and measurements, outputting normalized data in a structured
JSON format for estimation workflows.
● Legal Defense AI Platform:
A legal AI platform designed to help defendants and their advocates to fight smarter,
negotiate better, and win more. Built on a LangGraph-based RAG architecture with
Elasticsearch and Gemini integration, the system ingests multi-format legal documents,
provides AI-powered legal insights.
Its Core Infrastructure includes:
○ LangGraph-based RAG pipeline with Gemini integration, semantic chunking, and
advanced reranking.
○ Elasticsearch-based legal knowledge base indexing 15+ sources
○ Full GCP deployment with Django backend, Celery pipelines, and Docker
orchestration.
Advanced Computer Vision
I work with complex visual data, particularly satellite and real-world imagery.
● Real Estate Analytics Engine
This tool provides automated property analysis by combining satellite imagery,
street-level images, and AI-powered feature extraction.
The engine uses longitude and latitude coordinates to fetch satellite and street-view
images of a property. These images are then analyzed with Gemini (LLM), which
interprets visual data based on structured prompts to extract property features and
identify the property’s status. Roofs are segmented into GeoJSON polygons using a
fine-tuned SAM2 model.
The system can process properties individually by address or in bulk by ZIP code,
providing an extensive analysis.
Its main features are:
● Fetches satellite and street-view images using geolocation coordinates.
● Uses Gemini LLM to analyze images and extract detailed property features.
● Segments roofs into GeoJSON polygons with a fine-tuned SAM2 model.
● Supports property analysis by single address or ZIP code.
● Built with Django APIs, Celery, Docker, and deployed on GCP for scalability.
Tech: Gemini (LLM), SAM2 (roof segmentation), Django (API), Celery (asynchronous
tasks), Docker (containerization), GCP (deployment).
● Road Construction Site Analyzer: A tool that measures construction zones from noisy
satellite images, using CV techniques for denoising and segmentation and GPT-4 to
refine low-confidence results.
Core Tech Stack:
● AI / ML: OpenAI GPT-4, Google Gemini, RAG Systems, Vector Databases, Document
Intelligence (Google/Azure), PyTorch (SAM2), Great Expectations, OpenCV,
LangChain/Langgraph.
● Cloud & DevOps: GCP (BigQuery, Dataflow, Cloud Run, Pub/Sub), Azure, AWS,
Terraform, Docker, Kubernetes, GitHub Actions, Azure DevOps.
● Data & Backend: Python (Django, FastAPI, Celery), SQL, PostgreSQL, Redis.
● Frontend: Next.js, React.
