satyam = {
    "role"        : "Generative AI Engineer",
    "education"   : "B.Tech CSE @ IIITDM Jabalpur (2023–2027)",
    "location"    : "Jabalpur, India 🇮🇳",
    "specializes" : ["RAG Pipelines", "Agentic AI", "LLM Applications",
                     "Multimodal AI", "Real-Time Voice Agents"],
    "stack"       : ["Python", "FastAPI", "LangGraph", "LangChain",
                     "Groq", "FAISS", "Redis", "PostgreSQL", "Docker"],
    "achievements": [
        "🏆 Amazon ML Summer School 2025 — Top 3,000 / 100,000+ students",
        "🥇 Amazon ML Challenge — Top 500 / 9,000+ teams"
    ],
    "currently"   : "Building scalable AI systems & seeking AI/ML internships"
}

I don't just build models — I build complete, production-grade AI systems.

An end-to-end production AI voice system for banking — with safety built in.

Architecture highlights:

• 🔊 Streaming STT/TTS pipeline via Whisper — enables low-latency real-time conversation
• 🧠 Hybrid RAG — FAISS (vector) + BM25 (lexical) + Cross-Encoder reranking for high-quality, grounded answers
• 🛡️ Multi-layer safety — prompt-injection defense + confidence-based Human-in-the-Loop escalation
• 🔁 LangGraph agentic orchestration — deterministic routing with stateful multi-turn memory
• 🗄️ Scalable backend — Redis session memory + PostgreSQL logging + WebSocket streaming

Stack: Python · FastAPI · LangGraph · Groq (LLaMA 3 70B) · FAISS · BM25
       Whisper · Redis · PostgreSQL · WebSockets · React

Automates video compliance auditing using multimodal AI — STT + OCR + RAG + LLM reasoning.

Architecture highlights:

• 🎬 Multimodal pipeline — Azure Video Indexer extracts transcripts + OCR text from video frames
• 🔍 Hybrid semantic retrieval — vector + keyword + reranking over a compliance corpus (37 chunks, top-3 retrieval)
• ✅ Automated PASS/FAIL decisions — schema-constrained LLM outputs for reliable downstream consumption
• ⚙️ End-to-end latency: 145.8 sec/video — fully deployed FastAPI microservice with CLI + API support

Stack: Python · FastAPI · LangGraph · Groq (LLaMA 3.1) · Azure AI Search
       Azure Video Indexer · Sentence Transformers · OpenTelemetry

Multi-agent system that automates literature reviews — cutting research time by 60%+.

Architecture highlights:

• 🤝 Multi-agent orchestration via CrewAI + LangGraph — specialized agents for search, synthesis, and report writing
• 📦 Hybrid RAG over 1000+ embeddings using Pinecone for persistent vector storage
• 📝 Structured report generation with persistent session memory across research sessions
• ⚡ Reduced manual literature review time by 60%+

Stack: CrewAI · LangGraph · Pinecone · LLaMA 3 · RAG · SQLite · Python

Production ML system for symptom-to-diagnosis mapping — 88% accuracy, RAG-grounded.

Architecture highlights:

• 🩺 88% accuracy across 100+ medical conditions using Random Forest classifier
• 🔗 95% factual alignment via RAG + Gemini API grounding — reduces hallucinations
• 🔄 Automated retraining pipeline — model stays current without manual intervention
• 🐳 Fully containerized with Docker for reproducible, scalable deployment

Stack: Python · RandomForest · RAG · Gemini API · Flask · Docker · Scikit-learn

◆ Advanced RAG Architectures     — hybrid retrieval, reranking, adaptive chunking
◆ Agentic AI Systems             — multi-agent orchestration, tool use, planning
◆ MLOps & Production AI          — monitoring, evaluation, scalable deployment
◆ LLM Evaluation & Guardrails    — hallucination reduction, safety, HITL systems

Open to AI/ML internships & full-time roles in Generative AI / LLM Engineering