Mridul Jain

I'm an AI Engineer and researcher working across machine learning, causal AI, open source, and applied AI.

I've published two IEEE papers, including a first-author paper, and hold a patent from my undergraduate work. I was also selected for European Summer of Code 2025, contributing to pgmpy and sktime.

Currently at TCS, I build AI and agentic solutions for enterprise clients. I'm particularly interested in machine learning, causal AI, intelligent systems, and building AI that works in the real world.

Software Engineer at Tata Consultancy Services (TCS) (2025-01 – Present)

• Increased unstructured document-processing volume by 35% by integrating local LLM inference pipelines (Ollama) into enterprise document-intelligence workflows.
• Improved processing throughput by 40% and reduced request latency by architecting scalable, asynchronous REST API systems in FastAPI/Flask for high-concurrency enterprise traffic.
• Raised semantic retrieval precision by 15% and cut document lookup time to sub-100ms by engineering vector search pipelines (Qdrant) for high-performance document retrieval.
• Shortened deployment sprint cycles and improved code maintainability by applying system design principles to decompose a monolithic processing module into a modular, production-ready microservices architecture.

Core Developer at pgmpy (Open Source) (2025-01 – Present)

• Increased monthly active computational deployments by implementing advanced causal-discovery algorithms (Fast Causal Inference, MAGs, PAGs) for the library's analytics suite.
• Reduced algorithmic complexity for m-separation validation checks by architecting a system-level refactor of core graph-manipulation classes with strict type safety and modular, object-oriented inheritance design.
• Enabled 5+ core technical feature roadmaps and removed cross-functional development bottlenecks by leading code reviews and architectural governance as primary maintainer.
• Scaled global contributor onboarding (GSoC/ESoC) by mentoring and directing new open-source contributors across distributed engineering teams.

Open Source Contributor at sktime & SciML (Julia Ecosystem) (2024-01 – Present)

• Reduced cold-start microservices memory footprint by decoupling core package dependencies and implementing time-series forecasting models natively in PyTorch.
• Blended mechanistic physical constraints with data-driven neural networks by co-authoring hybrid scientific ML pipelines and optimizing differential equation solvers within the SciML ecosystem.

B.Tech in Information Technology – Vellore Institute of Technology (VIT) (2021-01 – 2025-12)