I'm Saurav Mukherjee, an AI Engineer at EY working on enterprise agents, multimodal retrieval, evaluation, and governance. Outside work, I build products that turn messy information into something people can inspect, trust, and use—and help developers find their way into open source.
My recurring questions: Where did the evidence come from? What can the agent actually do? Where does human judgment belong?
Six ways I'm exploring those questions, from research and product development to community building.
SwaPilot explores an on-device AI OS for business: local intelligence, business memory, and workflows. The public repository contains the pilot website, research hub, and interactive product demo. |
OpsWeave turns PDFs, images, audio, spreadsheets, and API specifications into versioned workflows—with typed validation, evaluation gates, human approvals, and inspectable execution. |
HealthAI Agent Lab is an educational health-screening research system combining open-weight language models, local voice inference, versioned ONNX specialists, and deterministic safety controls. |
CodeIN Community is the developer community I founded in 2021. It brings together learning programs, mentorship, contribution pathways, browser utilities, maintainer tools, TypeScript libraries, and a CLI. |
SignalForge joins live incident rooms, turns conversation into attributable evidence, and coordinates specialist responders. Consequential actions stay under the human incident commander's control. |
VerdaTrace connects environmental approvals, amendments, maps, and field evidence to source-cited obligations, spatial review, and human-approved reports in an evidence-first MVP. |
Good systems need good communities. I founded CodeIN Community to help people move from following tutorials to building, collaborating, and making their first contribution.
| Community members | Events hosted | Contributors credited | Cloud learners mentored |
|---|---|---|---|
| 15,000+ | 45+ | 150 | 4,000+ |
The first two figures are published CodeIN milestones; 150 is its preserved contributor baseline. The 4,000+ learners reflect my work with the Google Cloud Ready Facilitator Program.
- Maintaining in public: CodeIN's original history and contributor credits, alongside a modern website and open developer toolkit.
- Teaching and mentoring: former Google DSC Lead (2022), Google Cloud Ready facilitator, and 100+ live one-to-one coding classes through ConnectBud.
- Contributing beyond my projects: follow my pull requests for fixes and behavior-focused regression tests.
| Focus | Tools and practices |
|---|---|
| Agents & retrieval | Agentic RAG, multi-agent orchestration, MCP, Semantic Kernel, hybrid search, evaluations, guardrails |
| Applied AI | Azure AI Foundry, Azure OpenAI, Amazon Bedrock, Hugging Face, PyTorch, ONNX, scikit-learn |
| Product & platform | Python, TypeScript, FastAPI, Next.js, React, PostgreSQL, AWS, Azure, Docker, Terraform, CI/CD, observability |
At EY, I work on role-aware multimodal retrieval, grounded test-generation agents, and AI governance. Previously at SMIIT, I built ML-driven threat-detection pipelines for security telemetry. I hold a B.Tech in Computer Science and Engineering from Asansol Engineering College.
Visual credits & how this profile is built
Original animated 3D SVG hero, dot-matrix footer, social icons, and project cards, paired with Readme Typing SVG and Skill Icons. Inspired by the community's awesome profile READMEs. The main visuals are stored in this repository. Asset sources and credits →
I'm interested in applied AI, developer tools, open models, and products with a clear human purpose. Have a thoughtful problem, a collaboration idea, or a community initiative?
