Swastik Mukherjee - AI Engineer & Full Stack Developer

    Swastik
    Mukherjee

    AI Engineer & Full‑Stack Developer

    Building production-grade intelligent systems, local ML pipelines, and data-driven applications with real deployments.

    Portfolio
    Chennai, IN·
    Scroll
    © 2026

    I’m Swastik Mukherjee, a Full Stack and AI developer who builds intelligent, real-world applications at the intersection of software engineering and machine learning. I specialize in creating scalable products that move from idea to deployment.

    10+

    Projects

    3

    Year

    Learning

    Selected Work

    Projects

    Each built from scratch, shipped to real users, and maintained in production.

    3Projects
    Featured
    01
    Project 013 total

    HelixAI

    Local-first multi-agent orchestration system

    Desktop AI orchestrator that runs complex workflows without cloud dependency.

    • Decomposes user requests into DAG task workflows, routes to specialized workers via Redis Streams
    • Handles LLM inference (Ollama), tool execution, RAG lookups, and streaming outputs locally
    • FastAPI orchestrator + PostgreSQL state + Qdrant vector search, all running on your machine
    Tech
    ElectronReactTypeScriptFastAPIPythonRedis StreamsPostgreSQLQdrantOllama
    02
    Project 023 total

    FaceCheck.AI

    CNN classifier for real vs. AI-generated faces

    Real vs synthetic face detector trained on 180,000+ images with high-confidence classification.

    • Custom CNN architecture trained on real vs. AI-generated face dataset
    • FastAPI backend serves TensorFlow/Keras inference with confidence breakdowns
    • Clear API boundaries between frontend and ML backend, containerized with Docker
    Tech
    Next.jsFastAPITensorFlowKerasDockerTypeScript
    03
    Project 033 total

    AI-Driven Cognitive Warfare Simulator

    Misinformation analysis pipeline: generation, propagation, detection, XAI

    End-to-end platform that generates or ingests narratives, simulates how they spread through a social graph, scores risk with a trained GNN, and explains the score with SHAP.

    • Graph Attention Network (GraphGATClassifier) trained offline on FakeNewsNet,LIAR and PHEME, exported for runtime inference
    • Simulates multi-step exposure spread over a synthetic graph, extracting velocity, bot ratio, echo density, and cascade depth
    • SHAP-based explanation engine turns the four-feature runtime vector into ranked drivers, a risk tier, and a plain-language recommendation
    • Next.js analyst UI with a live news-triage flow (NewsAPI/GDELT) and a three.js graph visualization tied to propagation metrics
    Tech
    Next.jsTypeScriptFastAPIPyTorchGNNSHAPthree.jsOllama

    I'm best at wiring up AI models to real products—from data to model to UI.

    Technical Stack

    Languages

    Core programming languages I use daily in production.

    Core
    TypeScriptPythonJavaScriptCC++
    Working Knowledge
    SQLGoJavaRustBash

    Full-Stack

    Frameworks and tools I reach for when building web apps.

    Core
    Next.jsReactNode.jsExpressFastAPIMongoDB
    Working Knowledge
    PostgreSQLTailwind CSSElectronFramer MotionZustandFlutter

    AI / ML

    Core frameworks and libraries I use for building, training, and deploying ML systems and workflows.

    Core
    PyTorchTensorFlowScikit-learnLangChain
    Working Knowledge
    TransformersKerasXGBoostRAG frameworksOllamaQdrantONNXSentence-TransformersHugging Face

    Tools & Infrastructure

    Dev tooling and infrastructure I rely on for building routes to production.

    Core
    DockerGitHubGitVS CodeLinuxVercel
    Working Knowledge
    RedisSupabaseFirebaseGitHub ActionsRenderGrafanaPrometheusPostman

    Contact

    Looking for an AI or full‑stack intern?

    2nd-year B.Tech student. I build real systems, ship working code, and own features end-to-end. Available for internships.

    Also on