knowledge-base.js

// ─── Knowledge Base ───
// Sourced from career-garage Resume-Factory (Sep 2026)

const knowledgeBase = {
  about: {
    name: 'Arijit Kumar Roy',
    role: 'Senior Software Engineer, Technical Lead — Data & AI Platform Engineering',
    headline: 'Data & Strategy Lead for Voice AI / Voice Agents',
    experience: '8 years',
    location: 'Bangalore, India',
    email: 'arijitroy003@gmail.com',
    github: 'arijitroy003',
    linkedin: 'sudo-kill',
    website: 'https://arijitroy003.github.io'
  },
  experience: [
    {
      company: 'Red Hat',
      role: 'Senior Software Engineer, Technical Lead — Data & AI Platform Engineering',
      period: 'Apr 2024 — Present',
      highlights: [
        'First hire of the Data & AI Platform team; leading 5 engineers on production agents',
        'On-Call / Data Reliability Agent (MCP, LangChain, Langfuse) cut investigation MTTR from ~38 minutes to ~4 minutes',
        'Data contracts and catalog completeness for 150+ data products; adoption grew from ~30 to 730 users',
        'AI-assisted GitLab merge-request reviewer on Vertex AI with LLM evaluation',
        'Release Assistant saved 1000+ lead engineer hours annually',
        'GitOps data plane on OpenShift (Python, Golang, Kubernetes); $200k+ platform cost reduction from legacy Redshift/Starburst'
      ]
    },
    {
      company: 'Beem',
      role: 'Senior Data Engineer — Financial Services',
      period: 'Nov 2023 — Mar 2024',
      highlights: [
        'LLM-powered data and AI platform serving 50M+ users for personal finance',
        'Data insights for investor pitches that secured $16k Databricks funding with $24k in future commitments',
        '500 GB/day clickstream pipelines on AWS S3, Python, Databricks, and Mixpanel'
      ]
    },
    {
      company: 'Tata Digital (Tata Neu)',
      role: 'Senior Software Engineer — E-Commerce, Retail',
      period: 'Aug 2021 — Oct 2023',
      highlights: [
        'Conversational AI (voice and text) platform serving 120M+ users and 500M events/day',
        'Voice-call analysis with speech-to-text, audio ML, and custom voice models',
        'Monitored an AI chatbot across 12 Indic languages',
        'Quality metrics: unique/repeat users, response time, intent and escalation funnels; 24x7 Kusto dashboards',
        'Led 6 engineers on 15+ customer-facing pipelines; lakehouse migration cut latency 75% and cost 80%',
        'GenAI product search with Azure OpenAI, LangChain, embeddings, and vector databases'
      ]
    },
    {
      company: 'Gnosis Lab',
      role: 'Founding Engineer — NASSCOM 10K Startups',
      period: 'Jun 2019 — May 2021',
      highlights: [
        'Founding engineer for 0-to-1 SaaS backends on AWS',
        'Social-marketing automation, an AI marketing bot, and an LMS delivering 50+ APIs'
      ]
    }
  ],
  skills: {
    voice: ['Conversational AI', 'STT', 'Audio ML', 'Multilingual (12 Indic)', 'Call analytics', 'Intent / escalation funnels'],
    agents: ['LangChain', 'MCP', 'Langfuse', 'LLM-as-judge', 'Tool / function calling', 'Cursor', 'Claude Code'],
    data: ['SQL', 'Data quality', 'Data contracts', 'Metric definitions', 'Collection pipelines', 'Catalog / semantic layer', 'Atlan', 'Snowflake', 'Databricks', 'PySpark', 'Delta Lake', 'dbt', 'Airflow', 'Kafka', 'Fivetran'],
    platform: ['Python', 'Golang', 'Kubernetes', 'OpenShift', 'GitOps', 'AWS', 'Azure', 'GCP', 'Observability'],
    languages: ['Python', 'Golang', 'SQL', 'JavaScript', 'TypeScript', 'Scala', 'Shell'],
    spoken: ['English', 'Bengali', 'Hindi']
  },
  education: [
    { school: 'Jadavpur University', degree: 'Masters in Computer Applications', focus: 'Distributed Systems / Cloud Computing', gpa: '8.81/10', period: '2018-2021' },
    { school: 'Ramakrishna Mission Vidyamandira', degree: 'B.Sc. Computer Science', gpa: '8.49/10', period: '2015-2018' }
  ],
  projects: [
    'On-Call / Data Reliability Agent at Red Hat (MCP, LangChain, Langfuse)',
    'GitOps Data Mesh and catalog plane at Red Hat (150+ data products)',
    'Conversational AI voice and text platform at Tata Neu (120M+ users)',
    'Voice-call analysis and 12-language chatbot monitoring at Tata Neu',
    'GenAI product search at Tata Neu',
    'LLM-powered personal finance platform at Beem',
    'Doctor — clinical management app (React Native, GCP, Firebase)'
  ]
};

const responses = {
  greeting: [
    'Hello. I am Arijit Kumar Roy’s portfolio assistant. I can walk you through his experience, projects, and how to get in touch.',
    'Welcome. I can help you learn about Arijit’s work in voice AI, data platforms, and agentic systems. What would you like to start with?'
  ],
  experience: () => {
    const exp = knowledgeBase.experience;
    return (
      `Arijit has **${knowledgeBase.about.experience}** of experience across voice AI, data platforms, and production agents.\n\n` +
      `He is currently **${exp[0].role}** at **${exp[0].company}** — first hire of the Data & AI Platform team, leading five engineers.\n\n` +
      `Before that: **${exp[1].company}** (LLM finance platform, 50M+ users), **${exp[2].company}** (conversational AI for 120M+ users), and founding engineer at **${exp[3].company}** (NASSCOM 10K).`
    );
  },
  redhat: () => (
    `**Red Hat** (Apr 2024 — Present) — Senior Software Engineer, Technical Lead, Data & AI Platform Engineering.\n\n` +
      `• First hire of the team; leads 5 engineers on production agents\n` +
      `• On-Call / Data Reliability Agent (MCP, LangChain, Langfuse) reduced investigation MTTR from ~38 minutes to ~4 minutes\n` +
      `• Data contracts and catalog completeness for 150+ data products; adoption grew from ~30 to 730 users\n` +
      `• GitLab merge-request reviewer on Vertex AI with LLM evaluation\n` +
      `• Release Assistant saved 1000+ lead engineer hours a year\n` +
      `• GitOps data plane on OpenShift; $200k+ platform cost reduction from legacy Redshift/Starburst`
  ),
  tata: () => (
    `**Tata Digital / Tata Neu** (Aug 2021 — Oct 2023) — Senior Software Engineer, Strategic Initiatives.\n\n` +
      `• Conversational AI (voice and text) for **120M+ users**, **500M events/day**\n` +
      `• Voice-call analysis with speech-to-text, audio ML, and custom voice models\n` +
      `• Chatbot monitoring across **12 Indic languages**\n` +
      `• Quality metrics: unique/repeat users, response time, intent and escalation funnels; 24x7 Kusto dashboards\n` +
      `• Led 6 engineers; lakehouse migration cut latency **75%** and cost **80%**\n` +
      `• GenAI product search with Azure OpenAI, LangChain, embeddings, and vector databases`
  ),
  beem: () => (
    `**Beem** (Nov 2023 — Mar 2024) — Senior Data Engineer, financial services.\n\n` +
      `• LLM-powered data and AI platform serving **50M+ users** for personal finance\n` +
      `• Contributed data insights to investor pitches that secured **$16k** Databricks funding with **$24k** in future commitments\n` +
      `• Built **500 GB/day** clickstream pipelines on AWS S3, Python, Databricks, and Mixpanel`
  ),
  gnosis: () => (
    `**Gnosis Lab** (Jun 2019 — May 2021) — Founding Engineer, NASSCOM 10K Startups.\n\n` +
      `• 0-to-1 SaaS backends on AWS (Python, Lambda, DynamoDB, MongoDB)\n` +
      `• Social-marketing automation, an AI marketing bot, and an LMS with 50+ APIs\n` +
      `This is where he learned end-to-end ownership as a first technical hire.`
  ),
  skills: () => {
    const s = knowledgeBase.skills;
    return (
      `**Voice / speech:** ${s.voice.join(', ')}\n\n` +
      `**Agents / evals:** ${s.agents.join(', ')}\n\n` +
      `**Data & strategy:** ${s.data.slice(0, 8).join(', ')}\n\n` +
      `**Platform:** ${s.platform.join(', ')}\n\n` +
      `**Languages:** ${s.languages.join(', ')}`
    );
  },
  ai: () => (
    `Arijit’s AI work is production-facing, not demo-only.\n\n` +
      `• Conversational AI at Tata Neu: voice + text, 120M+ users, speech/STT/audio ML across 12 Indic languages\n` +
      `• Agents at Red Hat: MCP, LangChain, Langfuse, and LLM-as-judge evaluation\n` +
      `• On-Call / Data Reliability Agent cut MTTR from ~38 minutes to ~4 minutes\n` +
      `• GitLab MR reviewer on Vertex AI with scored LLM evaluation\n` +
      `• GenAI product search with embeddings and vector databases (Milvus, Chroma, Qdrant)\n\n` +
      `He works daily with Cursor and Claude Code.`
  ),
  projects: () => (
    `The work that usually matters most in conversation:\n\n` +
      `• **On-Call / Data Reliability Agent** at Red Hat — MCP, LangChain, Langfuse; MTTR ~38 min → ~4 min\n` +
      `• **GitOps data mesh + catalog** at Red Hat — 150+ data products, contracts, Atlan adoption 30 → 730 users\n` +
      `• **Conversational AI platform** at Tata Neu — voice and text, 120M+ users, 500M events/day\n` +
      `• **Voice-call analysis** — STT, audio ML, custom voice models, 12 Indic languages\n` +
      `• **GenAI product search** at Tata Neu — Azure OpenAI, LangChain, vector databases\n` +
      `• **LLM finance platform** at Beem — 50M+ users`
  ),
  education: () => {
    const edu = knowledgeBase.education;
    return (
      `• **${edu[0].school}** — ${edu[0].degree}\n  ${edu[0].focus}. GPA ${edu[0].gpa}. ${edu[0].period}.\n\n` +
      `• **${edu[1].school}** — ${edu[1].degree}\n  GPA ${edu[1].gpa}. ${edu[1].period}.\n\n` +
      `He also did information-retrieval research at ISI Kolkata under Dr. Dwaipayan Roy.`
    );
  },
  contact: () => {
    const a = knowledgeBase.about;
    return (
      `• **Email:** ${a.email}\n` +
      `• **GitHub:** github.com/${a.github}\n` +
      `• **LinkedIn:** linkedin.com/in/${a.linkedin}\n` +
      `• **Website:** ${a.website}\n` +
      `• **Location:** ${a.location}`
    );
  },
  location: () => 'Arijit is based in **Bangalore, India**. He is open to conversations about India-based roles and travel where it is useful.',
  hiring: () => (
    `Yes — he is open to the right next role.\n\n` +
      `Strongest fit: Voice AI / Voice Agents, data & strategy for model quality, agent evals, and platform engineering that production agents consume.\n\n` +
      `Please write to **arijitroy003@gmail.com** or LinkedIn (**sudo-kill**). A résumé can be sent on request.`
  ),
  currentWork: () => (
    `At **Red Hat**, Arijit is Senior Software Engineer and Technical Lead for Data & AI Platform Engineering.\n\n` +
      `Current focus:\n` +
      `• Production agents (On-Call / Data Reliability Agent) with MCP, LangChain, and Langfuse\n` +
      `• Data contracts and catalog quality for 150+ data products\n` +
      `• LLM evaluation on an internal GitLab MR reviewer\n` +
      `• GitOps / OpenShift data plane that those agents consume\n\n` +
      `He was the first hire on the team and leads five engineers.`
  ),
  achievements: () => (
    `Selected results, all from shipped systems:\n\n` +
      `• Investigation MTTR **~38 min → ~4 min** via the On-Call / Data Reliability Agent\n` +
      `• **150+** data products; catalog adoption **~30 → 730** users\n` +
      `• **1000+** lead engineer hours saved with Release Assistant\n` +
      `• **$200k+** platform cost reduction at Red Hat\n` +
      `• Conversational AI for **120M+** users and **500M** events/day at Tata Neu\n` +
      `• Lakehouse migration: **75%** lower latency, **80%** lower cost\n` +
      `• Voice-call quality loop across **12 Indic languages**\n` +
      `• Beem: **50M+** users; Databricks funding support ($16k + $24k committed)`
  ),
  hobbies: () => (
    `Outside work he plays chess, writes technical notes, and spends time on Linux tooling. The professional through-line is the same: careful systems, clear writing, and iterative improvement.`
  ),
  snowflake: () => (
    `At Red Hat he used **Snowflake** as part of a GitOps data mesh on OpenShift, with dbt, contracts, and catalog quality gates. The platform migration away from legacy Redshift/Starburst delivered **$200k+** in cost reduction. For voice-agent work, the more relevant layer is the governed metadata and evals those warehouses now feed.`
  ),
  databricks: () => (
    `**Databricks** and **Delta Lake** show up in two places: the Tata Neu lakehouse migration (**75%** faster, **80%** cheaper) and Beem’s LLM finance platform plus 500 GB/day clickstream. He also supported a Databricks funding conversation at Beem.`
  ),
  dbt: () => (
    `**dbt** is part of the Red Hat GitOps data mesh: tested SQL, documentation, lineage, and contracts so producer changes fail at the boundary instead of in production.`
  ),
  kubernetes: () => (
    `He runs data and agent workloads on **Kubernetes** and **OpenShift** at Red Hat — GitOps deployments, CI, and the Python/Golang services that agents and evals consume.`
  ),
  airflow: () => (
    `**Airflow** orchestrates dbt, Fivetran, and custom jobs in the GitOps data mesh. The operational bar is the same as for agents: known schedules, visible failures, and quality checks before data reaches consumers.`
  ),
  langchain: () => (
    `**LangChain** is in production in two systems: GenAI product search at Tata Neu, and the On-Call / Data Reliability Agent at Red Hat (with MCP and Langfuse). He treats it as an orchestration layer with evals, not a prototype kit.`
  ),
  mcp: () => (
    `At Red Hat he builds **MCP** servers so agents can query catalogs, lineage, and operational tools under guardrails. The On-Call / Data Reliability Agent is the main example: tool-calling against governed metadata, with Langfuse traces.`
  ),
  vectordb: () => (
    `At Tata Neu he used **Milvus**, **Chroma**, and related stores for GenAI product search — embeddings plus retrieval for 120M+ users. Vector search is one piece of that quality loop, alongside intent and escalation metrics.`
  ),
  gitops: () => (
    `The Red Hat data plane is **GitOps** on OpenShift: infrastructure, pipelines, and product definitions live in Git. Release Assistant and CI are how that scales to 150+ data products without deploy-by-chat.`
  ),
  dataops: () => (
    `His DataOps practice is CI/CD for transformations, data contracts, catalog completeness, and pipeline observability — so quality issues surface before a stakeholder meeting, not during one.`
  ),
  leadership: () => (
    `Leadership experience:\n\n` +
      `• First hire of Red Hat Data & AI Platform; leads **5** engineers\n` +
      `• Led **6** engineers at Tata Neu on 15+ customer-facing pipelines\n` +
      `• Founding engineer at Gnosis Lab\n` +
      `• Enablement workshops and Agentic SDLC practices for internal teams\n\n` +
      `The pattern is high ownership: 0-to-1, then quality bars, then scale.`
  ),
  startup: () => (
    `**Gnosis Lab** (NASSCOM 10K) was the founding-engineer chapter: AWS backends, an AI marketing bot, and an LMS with 50+ APIs. That same first-hire ownership shows up later as the first Data & AI Platform hire at Red Hat.`
  ),
  industries: () => (
    `• **Enterprise / open source** — Red Hat, data platforms and production agents\n` +
      `• **Fintech** — Beem, personal finance for 50M+ users\n` +
      `• **E-commerce / retail** — Tata Neu conversational AI and product search\n` +
      `• **SaaS** — Gnosis Lab, 0-to-1 products\n\n` +
      `The repeating problem is quality under scale: voice, events, and metadata that agents can trust.`
  ),
  research: () => (
    `At **ISI Kolkata** (Information Retrieval Lab, Dr. Dwaipayan Roy) he built an offline document indexer — early search/context work over unstructured corpora. Formal degrees: MCA from Jadavpur University (GPA 8.81) and B.Sc. Computer Science from Ramakrishna Mission Vidyamandira (GPA 8.49).`
  ),
  whyData: () => (
    `He works in data and voice AI because production conversations and pipelines have measurable failure modes. The useful loop is: observe a break, define a quality bar, collect or constrain the data, then measure again. That is the same loop behind Tata Neu’s voice analytics and Red Hat’s agent evals.`
  ),
  aboutBot: () => (
    `This site has two assistants:\n\n` +
      `• **This text chat** — answers from a local knowledge base, with an optional in-browser LLM\n` +
      `• **The voice widget** — a LiveKit agent that can speak with you\n\n` +
      `Neither records conversations for later review. I only use the facts on this page.`
  ),
  blog: () => (
    `The **blog** tab has technical writing on data mesh, MCP, vector databases, and lakehouse work. Open-source and side projects include **usernaut** (Kubernetes operator), **gitlab-mr-reviewer**, and **genai-toolbox**.`
  ),
  resume: () => (
    `Please email **arijitroy003@gmail.com** or use LinkedIn (**sudo-kill**) and a current résumé will be sent. You can also review experience, projects, and contact details on this site.`
  ),
  name: () => (
    `This assistant represents **Arijit Kumar Roy**, a Data & AI platform lead in Bangalore with **${knowledgeBase.about.experience}** of experience.\n\n` +
      `Current role: ${knowledgeBase.about.role} at Red Hat. Earlier: Beem, Tata Digital / Tata Neu, and founding engineer at Gnosis Lab.`
  ),
  joke: [
    'I can keep this professional — but if you would like a light one: why did the eval suite join the call? To score the conversation before anyone else did.',
    'A short one: the pipeline did not fail. It raised a quality signal. That is the polite version.'
  ],
  thankyou: [
    'You are welcome. I can continue with experience, projects, or contact details whenever you are ready.',
    'Glad that helped. What else would you like to know?'
  ],
  fallback: [
    'I may not have that detail. I can speak to experience, skills, projects, voice AI work, or how to get in touch.',
    'That is outside this knowledge base. Try experience, Red Hat, Tata Neu, projects, or contact.'
  ]
};

function detectIntent(input) {
  const lower = input.toLowerCase();

  if (/^(hi|hello|hey|greetings|howdy|yo)\b/i.test(lower)) return 'greeting';
  if (/thank|thanks|thx|appreciated/i.test(lower)) return 'thankyou';
  if (/joke|funny|humor|laugh|make me laugh|tell me something funny/i.test(lower)) return 'joke';
  if (/who\s*(are\s*you|is\s*(this|arijit))|your\s*name|introduce/i.test(lower)) return 'name';
  if (/hiring|hire|job\s*opportunit|open\s*to|available|looking\s*for\s*(work|job|role)|recruit|opportunit|position/i.test(lower)) return 'hiring';
  if (/current(ly)?|now|these\s*days|working\s*on|doing\s*now|present/i.test(lower)) return 'currentWork';
  if (/red\s*hat/i.test(lower)) return 'redhat';
  if (/tata|neu\b/i.test(lower)) return 'tata';
  if (/beem/i.test(lower)) return 'beem';
  if (/gnosis/i.test(lower)) return 'gnosis';
  if (/\bsnowflake\b/i.test(lower)) return 'snowflake';
  if (/\bdatabricks\b|delta\s*lake|pyspark/i.test(lower)) return 'databricks';
  if (/\bdbt\b|data\s*build\s*tool/i.test(lower)) return 'dbt';
  if (/kubernetes|k8s|\bkube\b/i.test(lower)) return 'kubernetes';
  if (/airflow|orchestrat/i.test(lower)) return 'airflow';
  if (/langchain/i.test(lower)) return 'langchain';
  if (/\bmcp\b|model\s*context\s*protocol/i.test(lower)) return 'mcp';
  if (/vector\s*(db|database)|milvus|chroma|qdrant|embedding/i.test(lower)) return 'vectordb';
  if (/gitops/i.test(lower)) return 'gitops';
  if (/dataops/i.test(lower)) return 'dataops';
  if (/achievement|accomplish|impact|metric|numbers|savings|result/i.test(lower)) return 'achievements';
  if (/\bblog\b|article|medium|writing|side\s*project|repos?\b/i.test(lower)) return 'blog';
  if (/hobbi|interest|free\s*time|fun|outside\s*work|chess|linux/i.test(lower)) return 'hobbies';
  if (/lead|leader|team|manage|mentor/i.test(lower)) return 'leadership';
  if (/startup|founding|nasscom|entrepreneur/i.test(lower)) return 'startup';
  if (/industr|sector|domain|fintech|ecommerce|e-commerce|retail|saas/i.test(lower)) return 'industries';
  if (/research|isi|indian\s*statistical|academic|paper|publication/i.test(lower)) return 'research';
  if (/why\s*(data|ai|this\s*field|engineer)|passion|motivation|interested/i.test(lower)) return 'whyData';
  if (/this\s*(bot|chatbot|ai)|how\s*(do|does)\s*(this|you)\s*work|built\s*this/i.test(lower)) return 'aboutBot';
  if (/resume|cv\b|curriculum/i.test(lower)) return 'resume';
  if (/voice|speech|stt|indic|langfuse|eval/i.test(lower)) return 'ai';
  if (/experience|work\s*history|career|roles?|companies|background/i.test(lower)) return 'experience';
  if (/skills?|tech\s*stack|technologies|proficient|languages?|programming|expertise/i.test(lower)) return 'skills';
  if (/\bai\b|llm|ml\b|machine\s*learning|openai|gpt|gemini|claude|mistral|genai|generative/i.test(lower)) return 'ai';
  if (/projects?|built|portfolio|showcase/i.test(lower)) return 'projects';
  if (/education|degree|university|college|study|school|gpa|masters|bachelor/i.test(lower)) return 'education';
  if (/contact|email|linkedin|github|reach|connect|social/i.test(lower)) return 'contact';
  if (/location|where|based|city|live|country|india|bangalore/i.test(lower)) return 'location';
  if (/python|golang|\bgo\b|sql|javascript|typescript|scala|terraform|docker/i.test(lower)) return 'skills';
  if (/aws|azure|gcp|cloud/i.test(lower)) return 'skills';
  if (/kafka|fivetran|atlan|mongodb|dynamodb/i.test(lower)) return 'skills';

  return 'fallback';
}

function getResponse(intent) {
  const resp = responses[intent];
  if (typeof resp === 'function') return resp();
  if (Array.isArray(resp)) return resp[Math.floor(Math.random() * resp.length)];
  return responses.fallback[0];
}