Skip to content
View MukulRay1603's full-sized avatar
12 followers · 11 following

Highlights

  • Pro

Block or report MukulRay1603

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
MukulRay1603/README.md
Typing SVG

     



M.S. Applied Machine Learning · University of Maryland, College Park (May 2026)
Every project below has real eval metrics, public repos, and live deployments.

✦ Tech Arsenal

Domain Stack
GenAI & LLMs LangChain · LangGraph · LoRA / QLoRA · RAG · Agentic AI · Multi-Agent Systems · vLLM · Pinecone · Guardrails · Ragas
ML & DL PyTorch · TensorFlow · Reinforcement Learning · Scikit-learn · OpenCV · NumPy · Pandas
MLOps & Infra MLflow · FastAPI · Docker · AWS (EC2/S3) · Azure (ACR/ACI) · GitHub Actions CI/CD · ONNX · Unity Sentis · Ray RLlib · CUDA
Languages Python · SQL · C++ · C# · Bash

✦ Featured Projects

⚔️ Nemesis — Distributed RL Boss AI with Real-Time LLM Narration

Metric Result
Training PPO · 3-stage curriculum · 300K steps · reward −2.4 → 13.2
Distributed Scaling Ray RLlib · 1 → 8 workers · 2.6× wall-clock speedup
Runtime Inference ONNX → Unity Sentis · <2 ms/frame · 60 fps · zero Python runtime
LLM Narration Groq LLaMA-3.3-70B + NeMo Guardrails over TCP · 114–352 ms
Eval RL boss outlasts scripted FSM baseline by 85% (37s vs 20s avg, 50 episodes)
Phase Escalation All 3 difficulty phases reached in 100% of episodes

PPO Ray RLlib Unity ML-Agents ONNX Unity Sentis Groq NeMo Guardrails MLflow C#

🛡️ EmpathRAG — NLI Safety Guardrail & Emotion-Conditioned RAG

Metric Result
Crisis Detection DeBERTa-v3-base NLI on 232K pairs · 0.9629 recall · 30 adversarial probes in 6 attack categories
Explainability captum Integrated Gradients on every safety intercept · auditable token-level XAI
Retrieval 5-stage pipeline · 1.67M-vector FAISS index · RoBERTa + LoRA (0.7127 F1)
Ablation Emotion alignment 0.88 vs 0.30 BM25 baseline
Significance Wilcoxon p = 3.62 × 10⁻⁸ (Condition D vs A)

DeBERTa RoBERTa LoRA FAISS captum NLI Mistral 7B (GGUF) PyTorch

🔍 RECON — Agentic ML Research Navigator  ·  🚀 Live Demo

Metric Result
Architecture 4-agent LangGraph state machine (planner → retriever → critic → synthesizer) + retry loop
Critic Taxonomy 4 verdicts: PASS / STALE / CONTRADICTED / INSUFFICIENT
Staleness Detection Catches 52% superseded ML claims vs 0% single-pass RAG (130-question eval)
Decay Ablation Linear optimal (52%) vs no-decay (42%) vs log-decay (38%)
Position Accuracy 43.9% vs 32.3% baseline

LangGraph Groq LLaMA-3.3-70B Semantic Scholar API Tavily Gradio HuggingFace Spaces

🌳 Irminsul — Production LLMOps System  ·  🚀 Live Demo

Metric Result
Fine-tuning Llama 3.1 8B · QLoRA (rank 16, lr 2e-4) · 3 MLflow-tracked experiments
Best Checkpoint Semantic similarity 0.826 · ROUGE-L 0.466
Serving Stack FastAPI + LangChain + Pinecone RAG + guardrails + confidence-gated web fallback
Corpus Pipeline irminsul-corpus · 840 docs · 6,876 vectors · 3 trust tiers
Automation GitHub Actions weekly ingest → Pinecone upsert · zero manual intervention

QLoRA LangChain Pinecone FastAPI Docker MLflow HuggingFace Spaces GitHub Actions CI/CD

✦ Publication

📄 Voice-Driven Panoramic VR Generation — IEEE IDCIoT 2024
Speech → 360° image pipeline (Whisper · GPT-Neo · Stable Diffusion) · 2.3× throughput via attention slicing on 10 GB VRAM
🔗 DOI: 10.1109/IDCIoT59759.2024.10467441

✦ GitHub Activity

  

✦ Currently

🔭  Wrapping up Nemesis — LLM-as-judge eval + demo recording
🎓  Graduating May 2026 — STEM OPT eligible
🎯  Seeking: AI/ML Engineer · GenAI · Game AI · LLMOps · RL Engineer
🧗  Off-hours: climbing walls, Genshin theory-crafting, houseplants

✦ "The stars, the sky... it's all a gigantic hoax." — Mona, probably ✦

Pinned Loading

  1. Irminsul Irminsul Public

    Fine-tuned Llama 3.1 8B · QLoRA · RAG serving stack · FastAPI · Pinecone · Azure

    Python 6

  2. nemesis-boss-ai nemesis-boss-ai Public

    PPO-trained adaptive boss AI deployed in Unity via ONNX/Sentis with real-time LLM narration

    ShaderLab 6

  3. project-recon project-recon Public

    Multi-Agent Research Navigator - LangGraph + Semantic Scholar

    Python 6

  4. Empath-RAG Empath-RAG Public

    EmotionAware RAG

    Python 6

  5. irminsul-corpus irminsul-corpus Public

    scrape the irminsul's corpus. Teyvat's Lore - Genshin Impact

    Python 5

  6. MukulRay1603.github.io MukulRay1603.github.io Public

    Personal portfolio — AI/ML Engineer

    HTML 6