vaibhav_rai/ vadodara, in

Vaibhav Rai

Software Development Engineer | Backend & AI Systems

B.Tech IT, Parul University (2023–2027)

Backend-focused engineer who builds production systems and LLM-powered agents: a job-intelligence pipeline, a RAG-based codebase-intelligence engine, and a tool-controlling portfolio agent. Strong in Node.js/Express, Redis, MongoDB/MySQL, Docker and Azure CI/CD.

SPIDY · guest/agent control plane / rolenova • reposage
live actions & probes:
conversations may be logged to improve response quality
production systems

RoleNova

Autonomous Job Intelligence Agent
code/live

Scrapes 72 company ATS APIs (Greenhouse, Lever, Workday, SmartRecruiters) plus a live Telegram channel on a cron schedule. Ingested and AI-evaluated 16,000+ listings with ~99% filtered out, delivering one ranked daily digest. Backed by a 5-tier LLM fallback cascade (Cerebras → Groq → OpenRouter → DeepSeek → heuristic) and a Redis caching/distributed locking layer cutting repeat-evaluation latency from ~2,500ms to <2ms.

  • ATS API Fleet: Scheduled cron workers ingest active listings across 72 company ATS APIs (Greenhouse, Lever, Workday, SmartRecruiters) plus a live Telegram channel.
  • Pre-AI Filtering & Ingestion: Evaluated 16,000+ listings with ~99% filtered out into one ranked daily digest, cutting compute overhead before calling LLMs.
  • 5-Tier Fallback Cascade: Multi-provider redundancy routes around HTTP 429 rate limits across Cerebras, Groq, OpenRouter, DeepSeek, and local rule evaluators.
  • Cost-aware deployment: GitHub Actions runs scraping, AI evaluation, Telegram polling, and the scheduled digest; MongoDB Atlas stores durable state while the Render service hosts the dashboard and API.
Node.jsMongoDB AtlasMongoDB5-Tier LLM CascadeDockerGitHub Actions

RepoSage

AI Codebase Intelligence Engine
code/live

Tree-sitter WASM + Python AST parsing into syntactically complete units without breaking scopes; Git history mining with LLM-synthesized intent. Dual-collection RAG in ChromaDB (live code + Git diffs, queried in parallel) with a two-tier offline semantic cache (Redis exact-match + local ONNX cosine search). Streams ingestion progress over SSE, answers with exact code citations. Deployed on Azure Container Apps with Redis caching and Docker Compose.

  • Dual-Collection Vector Segregation: ChromaDB segregates live code and Git diffs (queried in parallel) to eliminate semantic noise between source syntax and commit history.
  • Tree-Sitter WASM AST Parsing: Discovers files, streams SSE progress, and extracts syntactically complete function, class, and dependency units across JavaScript, TypeScript, and Python without arbitrary line slicing.
  • Git History Mining: Analyzes historical git diffs and commit deltas to infer engineering intent and architectural decisions behind codebase changes.
  • Two-Tier Offline Semantic Cache: Redis exact-match paired with local ONNX cosine search; deployed on Azure Container Apps with Redis caching and Docker Compose.
horizontal dataflow view
RepositoryJS / TS / Python+ Git Log DiffsAST ParsingTree-Sitter WASMGit ArchaeologyIntent AnalysisChromaDBreposage_codereposage_diffsParallel QueryRAG QueryFastAPI EngineExact CitationsSemantic Cache
FastAPIPythonChromaDBTree-Sitter WASMRedisDocker ComposeAzure Container Apps

Spider AI (“Spidy”)

Autonomous Portfolio Agent & Systems Control Plane
code

Node.js runtime with an isolated Python sidecar and ChromaDB RAG. Multi-LLM cascade (Groq → OpenRouter, with Gemini primary), SSE streaming, per-IP rate limiting, and off-topic guardrails. An internal intent dispatcher operates live production systems (RoleNova, RepoSage) through /api/agent/* endpoints, with destructive actions gated by a server-side session passcode. The public UI is deployed on GitHub Pages, while the private API and Python RAG sidecar run together on Render.

  • RAG & Sidecar Architecture: Node.js HTTP server paired with a Python ChromaDB RAG service running on port 8008, retrieving Top-5 semantic chunks with intent boosting.
  • Dual-Mode Security Model: Enforces read-only guest exploration for recruiters and an authenticated Jarvis-like owner mode unlocked via SHA-256 session passcodes.
  • Live Systems Dispatcher: Interacts with live production systems (RoleNova scraper, RepoSage AST RAG) via authenticated /api/agent/* routes.
  • Resilience & Streaming: SSE word-by-word streaming cadence, multi-tier fallback cascade, and strict off-topic guardrails.
Node.jsPythonChromaDBAI AgentsLangChainSSERenderGitHub Actions CI/CD
technical skills
  • Languages: JavaScript, Python, C++, SQL
  • Backend & Web: Node.js, Express.js, REST APIs, JWT, RBAC/PBAC, FastAPI, React
  • AI/LLM: RAG, LangChain, ChromaDB, Tree-sitter (AST), AI Agents, LLM APIs (Gemini, Groq, Cerebras, OpenRouter)
  • Databases: MongoDB, MySQL (indexing), Redis
  • DevOps: Docker, Azure, GitHub Actions CI/CD, Git
experience
Network18 / Backend Developer (Contract)
Aug 2025 – Dec 2025

REST APIs across modules of a production web app; multi-device JWT auth with refresh/access tokens, rotation and revocation; RBAC/PBAC middleware; MySQL schema refactor with composite indexes and automated data workflows, ~90% query efficiency gain.

Node.jsExpress.jsREST APIsJWT (Rotation & Revocation)RBAC/PBACMySQL Optimization (~90%)
certifications
  • SQL — HackerRank
  • Problem Solving — HackerRank
  • Java OOP — Internshala
  • Computer Networks — NPTEL
  • AWS Cloud Foundations — Amazon Web Services
algorithmic benchmarks live telemetry
1,000+
Problems Solved
LeetCode: 697+ | CodeChef | Algorithmic Rigor
Top 10.2%
LeetCode Global Standing
Rating 1761 / 697+ Solved
Rank 558
CodeChef Starters 227
Rating 1412 / Global Rank 41955
Top 5
Odoo Hackathon Finalist
Selected from 100+ teams