# Harsh Bafna **Machine Learning Engineer** · Bengaluru, India Machine learning engineer specializing in generative AI, multi-agent systems, and production-grade LLM applications. I build retrieval and analytics systems that turn messy enterprise data into useful answers. --- ## At a glance - **Role:** Machine Learning Engineer - **Location:** Bengaluru, India - **Email:** [harshbafna29@gmail.com](mailto:harshbafna29@gmail.com) - **Links:** [Portfolio](https://bafnaaharsh.netlify.app/) · [GitHub](https://github.com/bafnaharsh) · [LinkedIn](https://linkedin.com/in/bafnaaharsh) · [Resume (PDF)](https://bafnaaharsh.netlify.app/HarshBafna.pdf) - **Focus:** Machine Learning, Generative AI, Large Language Models, Retrieval-Augmented Generation, Multi-Agent Systems --- ## Summary Machine Learning Engineer with ~3 years of experience specializing in Generative AI, multi-agent orchestration, MLOps, and enterprise cloud architecture on GCP. Proven track record of architecting scalable retrieval systems (RAG), and deploying microservices-backed production AI platforms for enterprise data analytics. I am currently a **Machine Learning Engineer** at [Quantiphi](https://quantiphi.com/), where I build generative AI, RAG, and multi-agent systems for production enterprise workflows. Previously, I worked with [JP Morgan Chase & Co.](https://www.jpmorganchase.com/) I like systems that feel simple on the surface but do serious work underneath: retrieval pipelines, natural-language analytics, and AI tools that make data easier to use. **Here are some technologies I have been working with:** `Python` `SQL` `LangChain` `Google ADK` `Vertex AI` `FastAPI` --- ## Experience ### Machine Learning Engineer — [Quantiphi](https://quantiphi.com/) *FEB 2024 – PRESENT · Bengaluru, India* - Led 5+ ML engineers in a 24+ member team to ship production multi-agent systems for natural-language enterprise analytics. - Architected ADK-based agents with RAG over 14+ structured data sources, including Kafka topics, Google Ads, and Google Analytics schemas. - Built AI search, recommendation, campaign-generation, and lead-hotspot workflows that reduced analysis and drafting cycles from days to minutes. ### Software Engineer Intern — [JP Morgan Chase & Co.](https://www.jpmorganchase.com/) *NOV 2022 – SEP 2023 · Remote* - Designed a real-time data streaming interface capable of processing 100,000+ data points per second. - Used JP Morgan Chase's Perspective visualization tool to build live financial correlation graphs. - Implemented upper and lower-bound indicators for real-time monitoring of stock-correlation metrics. **Certificate:** [JP Morgan Forage Internship](https://bafnaaharsh.netlify.app/certificate/jp-morgan-forage-internship) --- ## Projects ### RCA, Segmentation & NL2SQL Bot *Quantiphi · AI Agents · ADK · BigQuery · Apache Kafka · RAG · Multi Agent Orchestration* Production multi-agent analytics platform that lets business users query structured enterprise data through natural-language chat. - Led a team of 5+ ML engineers within a 24+ member engineering group to design and ship 3 production-grade multi-agent orchestration systems, enabling executive teams to query enterprise metrics via natural language. - Architected an agentic analytics platform utilizing Google Agent Development Kit (ADK) with a custom RAG retrieval engine indexing schemas across 14+ structured data sources (Kafka topics, Google Ads, BigQuery) for precise NL2SQL synthesis. - Cut executive and management data-reporting turnaround time from days to under 2 minutes, accelerating decision-making velocity across business units. **Stack:** `Python` `SQL` `Google ADK` `LLMs` `RAG` `AI Agents` ### AI Search & Recommendation Engine *Quantiphi · Vertex AI Search · Dynamic Model Swap · Ranking · Keyword Extraction* Conversational product discovery system over an 80K+ SKU catalog, combining Vertex AI Search with recommendation models. - Architected a real-time conversational search and recommendation engine, replacing legacy fuzzy keyword matching with intent-aware semantic vector retrieval. - Engineered a high-throughput search platform using Vertex AI Search for Commerce, indexing a 1.5M+ SKU product catalog and implementing custom ranking models (frequently-bought-together, similar items, personalized scoring). - Increased catalog discovery accuracy and user engagement by aligning free-form natural language queries directly to product vector spaces. **Stack:** `Python` `Vertex AI Search` `LLMs` `Ranking` ### AI Email Campaign Generator *Quantiphi · Agentic Workflows · Imagen · Cloud Run · Web Scraping* AI workflow that scrapes property themes and metadata, then generates polished marketing emails with enhanced imagery. - Built an automated multimodal marketing workflow that extracts brand aesthetics and metadata from target web domains to generate custom email assets. - Engineered headless scraping pipelines (Playwright) to extract design assets, feeding structured contextual prompts into Gemini and Imagen models deployed on Cloud Run for dynamic asset generation. - Reduced marketing campaign creation cycles from weeks to under 5 minutes by completely automating asset extraction and copy generation. **Stack:** `Python` `LLMs` `Web Scraping` `NLP` `Imagen` ### Lead Hotspot Detection *Quantiphi · Hybrid Predictive Model · SDWIS Database · Geospatial Feature Engineering* Hybrid ML and rules-based system for predicting lead-contamination hotspots from SDWIS public water system data. - Engineered machine learning pipelines to predict municipal lead contamination risks across regional drinking water systems. - Preprocessed and engineered spatial features from Public Water System (PWS) datasets sourced from the EPA SDWIS database, implementing hybrid XGBoost and heuristic rule-based decision trees. - Achieved 88+% predictive accuracy for lead hotspot identification, directly optimizing resource allocation for infrastructure remediation teams. **Stack:** `Python` `LangChain` `LLMs` `EDA` `Feature Engineering` ### Real-Time Market Visualization *JP Morgan Chase & Co.* Streaming interface and live graphing workflow for monitoring historical stock-correlation behavior at high throughput. - Engineered a real-time financial streaming interface capable of ingesting and processing over 100,000 data points/second with minimal latency. - Utilized JP Morgan's Perspective open-source data engine to construct dynamic visualization dashboards monitoring historical stock pair correlations. - Implemented automated bound-crossing alert triggers based on statistical correlation metrics to support real-time quantitative monitoring. **Stack:** `Perspective` `Real-Time Data Streaming` `Visualization` --- ## Skills - **Programming & Core CS:** Python, C/C++, Java, SQL, Bash / Shell Scripting, Object-Oriented Programming (OOP) - **Generative AI & Agentic Systems:** Google Agent Development Kit (ADK), LLMs, Multi-Agent Orchestration, Retrieval-Augmented Generation (RAG), Model Context Protocol (MCP), LangChain, LangGraph, Prompt Engineering, Vector Search, Hybrid Search, Embeddings, Fine-Tuning, PEFT/LoRA, AI Guardrails, Nvidia NeMo - **Machine Learning & Data Science:** Neural Networks, PyTorch, TensorFlow, Scikit-Learn, XGBoost, LightGBM, Pandas, NumPy, OpenCV, Natural Language Processing (NLP), Computer Vision, Feature Engineering, Exploratory Data Analysis (EDA), Time-Series Forecasting - **GCP, Cloud & Infrastructure:** Google Cloud Platform (GCP), Vertex AI, Search/Recommendation Models, Cloud Run/AWS Lambda, Cloud Functions, Google Cloud Storage (GCS), Firebase, Pub/Sub, AWS, Azure, Docker, Kubernetes, DevOps, Linux - **Data Engineering & Architecture:** BigQuery, Data Warehousing, Apache Kafka, Snowflake, PostgreSQL, MongoDB, MySQL, Redis, NoSQL, Vector Databases, Data Pipelines, Microservices, REST APIs, FastAPI, Backend Architecture, Full-Stack Integration, System Design - **MLOps, Web & Tools:** CI/CD, MLflow, Hugging Face, Model Deployment, Streamlit, Git, GitHub, Document AI, Agile Methodologies --- ## Education ### B.M.S. College of Engineering *2020 – 2024 · Bangalore, India* Bachelor of Engineering, Electronics & Instrumentation Engineering — GPA: 8.3/10 ### S.Tech IT School *2019 – 2020 · Rajasthan, India* Higher Secondary School Certificate, PCM — Percentage: 82.6% --- ## Certifications - [Associate Cloud Engineer](https://bafnaaharsh.netlify.app/certificate/google-cloud-engineer) — Google Cloud - [Professional ML Engineer](https://bafnaaharsh.netlify.app/certificate/google-ml-engineer) — Google Cloud - [JP Morgan Forage Internship](https://bafnaaharsh.netlify.app/certificate/jp-morgan-forage-internship) — JP Morgan Chase & Co. - [One for All](https://bafnaaharsh.netlify.app/certificate/one-for-all) — Quantiphi - [Think Tank](https://bafnaaharsh.netlify.app/certificate/think-tank) — Quantiphi - [Consider It Done](https://bafnaaharsh.netlify.app/certificate/consider-it-done) — Quantiphi --- ## Links - [Portfolio website](https://bafnaaharsh.netlify.app/) — the human-readable version of this document - [GitHub profile](https://github.com/bafnaharsh) — code and experiments - [LinkedIn profile](https://linkedin.com/in/bafnaaharsh) — work history and contact - [Resume (PDF)](https://bafnaaharsh.netlify.app/HarshBafna.pdf) — one-page resume - [Email Harsh Bafna](mailto:harshbafna29@gmail.com) — harshbafna29@gmail.com