^{Metrics from production work at IQVIA (agentic AI for ROM/estimation) and DocMind's 13-question retrieval eval harness.}

frontend    react · redux-saga · mui data grid · react router
backend     node.js · express · .net 8 · fastapi
agentic-ai  langchain · tool-calling agents · vector re-ranking · groq/llama
data        mongodb · atlas vector search · sql server
cloud       aws (ec2 · s3 · lambda · iam · cloudwatch) · docker · firebase
quality     jwt/oidc · swagger/openapi · jest · gitlab api v4 ci/cd

react node/express python · fastapi langchain mongodb atlas

A tool-calling RAG agent that grounds itself in your documents first, and only reaches for outside help when it has to.

flowchart LR
    U([User Query]) --> A{Agent Loop}
    A -- "1: retrieve" --> D[(MongoDB Atlas<br/>Vector Search)]
    D -- "re-ranked context<br/>cross-encoder" --> A
    A -- "2: fallback" --> W[[Web Search<br/>Tavily]]
    A -- "3: fallback" --> C[[Sandboxed<br/>Calculator]]
    W --> A
    C --> A
    A --> G([Groq / Llama<br/>Generation])
    G --> R([Grounded Answer<br/>+ Citations + Trace])

    style A fill:#04241c,stroke:#00e676,color:#f4f6fb
    style G fill:#04241c,stroke:#00c9a7,color:#f4f6fb
    style R fill:#04241c,stroke:#00e676,color:#f4f6fb

• Cross-encoder re-ranking on retrieval lifted answer accuracy 85% → 92% on a 13-question evaluation harness built to score retrieval + generation quality
• Full execution trace — which tool fired, why, with what result — surfaced in the UI, not buried in logs
• Diagnosed and resolved production issues (memory limits, rate limiting, Docker) across three independently deployed services on Render

react node.js express mongodb socket.io

A Fiverr-style freelancer–client marketplace built solo, end to end — listings, real-time chat, transactions, and reviews on one MERN codebase.

flowchart LR
    subgraph Client["React Frontend"]
        UI[Browse / Gig Listings]
        ChatUI[Real-time Chat]
        AuthUI[Sign Up / Login]
    end

    subgraph Server["Node.js + Express API"]
        AuthSvc[Auth Service]
        ListingSvc[Listings API]
        TxnSvc[Transactions API]
        ReviewSvc[Ratings & Reviews API]
        Socket[[Socket Layer]]
    end

    DB[(MongoDB)]

    AuthUI --> AuthSvc --> DB
    UI --> ListingSvc --> DB
    UI --> TxnSvc --> DB
    UI --> ReviewSvc --> DB
    ChatUI <--> Socket --> DB

    style Server fill:#04241c,stroke:#00e676,color:#f4f6fb
    style DB fill:#04241c,stroke:#00c9a7,color:#f4f6fb

• Designed the full MongoDB schema architecture from scratch — users, gigs, transactions, and reviews as independently queryable collections
• Built secure authentication, a real-time chat layer, and end-to-end transaction workflows (request → accept → pay → review)
• React.js frontend consuming a self-built Node.js/Express REST API; published and documented on GitHub for others to reference

_{🏆 On stage at the SIH 2022 Grand Finale, Software Edition — team win, ₹1,00,000 prize}

GitHub metrics (auto-refreshes every 6h via .github/workflows/metrics.yml)

_{Requires enabling the included lowlighter/metrics workflow and adding a METRICS_TOKEN secret (a classic PAT with repo + read:user) — see setup notes below.}

_{Generated by .github/workflows/snake.yml on push to output}