Welcome to my GitHub! I'm Suman 👋

I specialize in architecting Multi-Agent systems, stateful LLM workflows, and scalable deep learning pipelines. As a Dual Degree (B.Tech + M.Tech) student at IIT Kharagpur (CGPA: 8.12), my focus is on bridging theoretical algorithmic problem-solving with deterministic, production-grade AI systems.

• 🔭 I’m currently focused on: Multi-Agent Orchestration (LangGraph), Long-Term Memory Systems, and Agentic RAG.
• 🌱 I have expertise in: Deep Learning (TensorFlow/Keras), LangChain, and orchestrating complex ML pipelines.
• 🏆 Milestones: Secured All India Rank (AIR) 64 in GATE 2024 | Recipient of the Prof. J.P. Ghose Memorial Award.
• 👨‍🏫 Leadership: Guided 500+ aspiring engineers at AlgoZenith in Data Structures, Algorithms, and interview problem-solving.
• 📫 How to reach me: sumanraj4176@gmail.com

• Generative AI & LLMs:
  

• Machine Learning & Deep Learning:
  

• Programming Languages:
  

• Frameworks and Libraries:
  

• Platforms & Tools:
  

• Core Engineering & Fundamentals:
  

• Tech Stack: Python LangGraph LangChain OpenAI Tavily Wikipedia LangSmith
• Built a collaborative multi-agent research system in LangGraph that decomposes complex topics into specialized analyst perspectives with human-in-the-loop refinement.
• Implemented parallel information retrieval using Tavily and Wikipedia to synthesize domain context.
• Designed a map-reduce workflow to parallelize analyst interviews and consolidate findings into structured reports.

• Tech Stack: Python LangGraph LangChain OpenAI Pydantic Trustcall
• Built a stateful task-management agent that maintains short-term conversational state and structured long-term memory across interactions.
• Designed discrete Profile, To-Do, and Instruction memory schemas with Pydantic for deterministic memory extraction and updates.

• Tech Stack: Python TensorFlow Keras MobileNetV3 OpenCV
• Developed an end-to-end Presentation Attack Detection (PAD) system using MobileNetV3-Small transfer learning to classify fingerprint images as LIVE or SPOOF.
• Calibrated operational thresholds against a 3% target BPCER and evaluated held-out test performance across APCER, EER, and ROC-AUC metrics.

• Tech Stack: Python TensorFlow Keras Scikit-Learn Pandas NumPy
• Engineered an end-to-end credit risk prediction pipeline across 396K+ loan records with missing-value handling and categorical encoding.
• Trained a deep neural network to classify loan default risk, achieving an F1-score of 0.93 on the test set despite significant class imbalance.

• Published an open-source Python profiling package on PyPI designed for line-by-line function tracing.
• Measures runtime statistics across repeated executions to help developers isolate bottlenecks and reduce latency.

• Teaching Assistant | IIT Kharagpur (Aug 2023 – May 2025)
  • Assisted in conducting tutorials and evaluating assignments for undergraduate courses, facilitating doubt-clearing sessions on core analytical concepts.
• Mentor | AlgoZenith (Oct 2022 – May 2024)
  • Mentored 500+ students in Data Structures and Algorithms, guiding them through coding challenges and optimization techniques.