Hi, I'm Spandan Mukherjee
I build AI systems and the full-stack products around them — MCP servers, RAG pipelines, and the web apps that put them in front of users.
Last 12 months
17 projects started · Oct 2025 — Sep 2026
About Me
I build stateful AI systems — MCP servers, hybrid RAG, and agent orchestration with LangGraph and A2A — alongside the full-stack and cloud work that puts them in front of real users. My day-to-day spans internal platform engineering, large-scale ETL, and LLM observability, and the part I care about most is where a prototype has to survive real traffic. Outside of that I ship developer tooling, publish CLIs on npm, and mentor in open source.
Work Experience
- Built and deployed ETL pipelines on AWS EC2 (Python) to standardize data from 112 organizational sources, eliminating significant manual processing overhead.
- Processed and normalized 10 years (2015–2024) of large-scale institutional educational data across CSV formats; engineered field mappings for nested ZIP archives containing 11 CSVs each, with benchmark validation checks to ensure accuracy.
- Engineered a 3-layer PDF extraction pipeline using Docling and Gemma4 (via DeepInfra) to parse complex tables, malformed cells, and inconsistent layouts from large institutional research documents.
- Developed the core internal platform on Azure VMs for composing workflows from scratch over available internal tools, with Docker-containerized one-click deployment converting single or chained flows into production full-stack applications, plus an integrated AI code generator serving live, shareable previews with user-configurable access and usage limits.
- Instrumented platform-wide LLM usage and cost tracing with benchmark comparisons across multiple models, and shipped a bring-your-own-model app letting users register any model and make it production-ready as an AI backend.
- Engineered a custom meeting bot from scratch that joins Google Meet sessions reliably without being flagged as a potential risk by the platform.
- Built stateful AI systems from scratch using Hybrid RAG, A2A Agent SDK, LangGraph, and Mem0/OpenMemory with Auth0 and encryption-based auth layers.
- Developed MCP Servers and MCP UIs with internal agent scripts and endpoints following AI engineering best practices.
- Integrated Langfuse for LLM tracing and evaluation, and built internal observability dashboards for monitoring agent performance.
- Refactored and contributed to a client production system on FastAPI, Vue.js, Azure (Cosmos DB, Document Intelligence, AI Foundry) and C#.
- Managed Kubernetes deployments with Helm and implemented blue-green deployment strategies across multiple branch phases.
- Built backend scripts for data migration and image generation model integration via Imagen pipelines on Azure and OpenRouter.
- Streamlined blob storage and session management, reducing frontend load and improving platform response.
- Worked with Web Scraping MCP outputs for data pipeline workflows.
- Produced tutorials and sample projects to streamline developer onboarding and product adoption.
- Hosted virtual developer events and gathered user feedback to improve product UX and community engagement.
- Built Jivan Companion using Gemini with fine-tuning on health-specific use cases and voice interaction support.
- Translated Figma designs into responsive frontend and improved model responses through iterative user feedback testing.
- Automated repetitive workflows in CentOS using Shell scripts and Ansible playbooks for configuration management and deployment.
- Enhanced system reliability through CI/CD practices and reduced manual intervention in testing and environment setup.
Featured Projects
Agent infrastructure, developer tooling, and full-stack products — the work I'd point at first.
A voice and vision AI astrologer with local RAG that delivers Vedic astrology and palm-reading guidance from your birth chart and palm on camera. Features local FAISS indexing, real-time speech via Gemini Realtime, and nakshatra/dasha/compatibility analysis grounded in authoritative astrology books.
A CLI tool that generates production-ready MCP servers from docs, APIs, OpenAPI/GraphQL schemas, and local codebases. Supports universal LLM providers (Ollama, OpenAI, Anthropic, Groq, etc.) and integrates with Claude Desktop and Archestra with a single command.
A conversational data analysis app where you upload datasets and ask questions in natural language to get summaries, visualizations, SQL, and runnable Python code. Features a Jupyter notebook panel, MCP integration, and built-in ML/DA/DV analysis workflows.
- Enabled dataset uploads with AI-generated visualizations, SQL, and runnable Python via natural language.
- Containerized the Jupyter MCP runtime with Docker and deployed the Astro docs on Vercel.
An AI-powered pitch analysis system built on LangGraph and Streamlit, providing investment analysis, pitch evaluation, and success prediction from historical Shark Tank data across the US, India, and Australia.
- Built a LangGraph workflow with 8 agents for industry insights, shark profiling, and success prediction.
- Deployed a Random Forest model for real-time pitch scoring and investment recommendations.
An adversarial AI simulation platform for evaluating agent reasoning and robustness, running real-time multi-agent matches between Red Team and Blue Team agents with live state updates.
- Built a full-stack system enabling real-time multi-agent AI matches with live state updates.
- Implemented LLM integrations and core APIs for agent orchestration and performance evaluation.
An AI-powered newsletter platform delivering personalized entertainment and cultural insights, powered by Qloo's trend data and Google's Gemini AI. Generates 400–600 word articles from real-time entertainment trend signals.
Shipped by Year
Everything I started and put on GitHub, year by year. Pulled live from the API — it updates itself as I ship.
11 repositories shipped in 2026
Achievements & Awards
Cline Track Winner
WeMakeDevs Hackathon · 2025
Scroll Track Winner
ETHIndia · 2023
2nd Position
DEFY · 2024
Runners Up
Kavach Hackathon · 2023
Finalist
Gen AI Hackathon — Neurealm (formerly GAVS) · 2024
Beta Microsoft Learn Student Ambassador
Microsoft · 2024
Speaking & Writing
Invited Speaker ×4
College talks on UI/UX, tech innovation, and cloud computing models.
Hackathon Judge ×4
Evaluated submissions and mentored teams across four hackathons.
Local-First AI Astrologer — technical write-up
A deep-dive on building a private, multimodal Vedic AI agent with local RAG.
Read itDeveloper tutorials & sample projects
Written as Developer Advocate at Quira to streamline developer onboarding and product adoption.
Education
Vellore Institute of Technology, Chennai
B.Tech in Computer Science and Engineering
CGPA: 8.38 / 10
September 2021 - September 2025 · Chennai, India
St. Paul's H S School, Indore
Class XII — PCM with Physical Education
89.2%
April 2020 - April 2021 · Indore, India
Open Source & Community
Led an open source blockchain project through GSSoC, reviewed contributor pull requests, clarified project guidelines, and unblocked contributors on technical issues.
VisitManaged the Push Protocol India Ambassador program, planned and executed growth strategies, and contributed to maintaining the Push documentation.
VisitBuilt and grew the club's cloud computing culture on campus, prototyping ideas with a team-based system and running events to support the wider tech community.
Represented Nordek in developer communities and supported ecosystem growth.
Frontend development and site maintenance for the Hack Club VIT Chennai web presence.
Contact Me
Get In Touch
Feel free to reach out to me for collaboration, opportunities, or just to say hello! I'm always open to discussing new projects, creative ideas, or opportunities to be part of your vision.