Back to Work Drive
WORK / Projects

Projects

6 featured + 4 additional projects — 10 total

Featured Projects

01

Oriz — 1000+ Free Online Tools Platform

Featured
TypeScriptReactAstroPythonCloudflare WorkersFirebaseRazorpay
  • Engineered a production-grade full-stack platform (oriz.in) with 192+ client-side tools across 8 categories using Astro, React, TypeScript, and Tailwind CSS, deployed on Cloudflare Pages.
  • Built a real-time data API marketplace with 69 Python web scrapers across finance, crypto, weather, sports, and news domains, orchestrated by 5 GitHub Actions CI/CD pipelines with scheduled cron jobs.
  • Integrated 10 AI/LLM providers into a unified chatbot interface with provider-agnostic abstraction and intelligent fallback routing.
  • Architected multi-cloud backend: Firebase (Auth + Firestore), Supabase, Turso (LibSQL), Upstash Redis, Cloudflare R2, Algolia Search, Sanity CMS, and Razorpay payment gateway with webhook-driven order verification.
  • Implemented 40+ cryptographic hash algorithms, client-side encryption (AES/DES/3DES/RC4), Kinde Auth (PKCE), Firestore security rules, and 100% client-side processing for maximum user privacy.
  • Built 8 serverless edge functions on Cloudflare Workers handling comments, ratings, file uploads (R2), email dispatch, reCAPTCHA verification, payment webhooks, and view tracking.
  • Embedded an Agentic AI Assistant with tool-calling capabilities into the platform, enabling users to invoke any of the 192+ tools via natural language commands — the agent autonomously selects, chains, and executes the correct tools to fulfill multi-step user requests.
02

NexusAI - Multi-Agent RAG Platform

Featured
PythonLangGraphLangChainAutoGenCrewAIOpenAINeo4jDockerKubernetes
  • Architected a production-grade Multi-Agent AI Platform using LangGraph that orchestrates a hierarchy of specialized autonomous agents — a Planner Agent, Executor Agent, Critic/Reflection Agent, and Memory Agent — to collaboratively decompose, solve, verify, and store solutions for complex, long-horizon coding and research tasks.
  • Designed stateful agent graphs with conditional branching, loop-back reflection cycles, and human-in-the-loop (HITL) checkpoints, enabling agents to self-correct by critiquing their own outputs before delivering final results.
  • Implemented a Graph-based Agentic RAG pipeline using Neo4j and Vector Embeddings, allowing agents to perform multi-hop knowledge retrieval across interconnected enterprise documentation nodes, achieving 40% higher answer accuracy vs. naive RAG.
  • Engineered persistent agent memory with both short-term in-context buffers and long-term vector-store memory (Chroma), enabling agents to recall prior task states and user preferences across sessions.
  • Built a dynamic tool registry allowing agents to autonomously discover, select, and invoke tools (web search, code execution, file I/O, API calls) using function-calling interfaces, following the ReAct (Reason + Act) pattern.
  • Deployed the multi-agent system on Kubernetes with auto-scaling capabilities to handle concurrent agent execution threads, with full observability via LangSmith tracing and Prometheus metrics.
03

AgentForge - No-Code AI Agent Builder Platform

Featured
PythonLangChainCrewAIFastAPIReactDocker
  • Built a visual drag-and-drop platform enabling non-technical users to design, configure, and deploy custom AI Agents by composing agent roles, tools, memory types, and LLM backends through a React-based UI.
  • Engineered a CrewAI-based multi-agent orchestration engine on the backend, supporting role-based agents (Researcher, Writer, Coder, QA, Manager) with configurable collaboration modes (sequential, hierarchical, parallel).
  • Developed a universal tool-use adapter supporting 25+ built-in tools (Google Search, Wikipedia, Python REPL, File System, REST API caller, SQL executor, Email sender) that agents can dynamically invoke during task execution.
  • Implemented long-term agent memory with automatic summarization and retrieval via Chroma vector store and Redis caching, enabling context persistence across multi-session workflows.
  • Exposed the platform via a FastAPI backend with real-time agent execution streaming (SSE), comprehensive run logs, token usage analytics, and exportable agent configurations as reusable YAML templates.
04

BrowserPilot - Autonomous Browser Agent

Featured
PythonPlaywrightLangChainGPT-4oFastAPI
  • Engineered a fully autonomous Browser-Use AI Agent powered by GPT-4o vision and Playwright, capable of navigating websites, filling forms, extracting structured data, and completing multi-step web tasks from plain English instructions.
  • Implemented a vision-grounded action loop where the agent takes a screenshot, uses a multimodal LLM to interpret the DOM and decide the next action (click, type, scroll, navigate), executes it, and repeats — achieving goal-directed browsing with no human intervention.
  • Built a task decomposition planner that breaks complex instructions into atomic sub-tasks, manages state across steps, and handles dynamic page changes gracefully.
  • Integrated anti-hallucination grounding: the agent verifies each action against live DOM state before committing, and uses a Critic sub-agent to validate task completion criteria before marking tasks as done.
05

TubeDigest - Multimodal Sponsor Detection

Featured
PythonTransformersONNX RuntimePyTorch
  • Developed a high-performance AI engine to detect sponsor segments using Hugging Face Transformers and multimodal analysis (Audio + Text).
  • Fine-tuned a T5 Model and optimized inference speed by 3x using ONNX Runtime and dynamic quantization techniques.
  • Built a scalable Flask microservice to process real-time video streams with sub-second latency, incorporating secure API endpoints.
  • Integrated a LangChain-based summarization agent that autonomously generates chapter summaries, key takeaways, and timestamps post-detection — turning raw transcripts into structured, actionable content briefs.
06

Olivia - Edge AI Voice Assistant with Agentic Capabilities

Featured
PythonEdge AILlama-3TensorFlow LiteLangChain
  • Engineered a privacy-first Virtual Assistant utilizing Local LLMs (Llama-3) for intent classification and complex command execution, running entirely on-device with no cloud dependency.
  • Implemented an Edge Computing architecture to process voice commands locally, eliminating cloud latency and ensuring offline functionality.
  • Designed a plugin-based Agentic Tool System enabling the assistant to autonomously invoke IoT device controls, system automation scripts, calendar management, and web lookups as contextual tools.
  • Implemented persistent conversational memory with session summarization and episodic recall, allowing Olivia to remember user preferences, prior commands, and long-running tasks across power cycles.

More Projects

Crawl4AI - Distributed RAG Ingestion

PythonRedisDistributed CrawlingSeleniumLangChain
  • Designed a Distributed Web Crawler utilizing Redis Task Queues to ingest massive datasets for LLM training and Agentic RAG pipelines.
  • Implemented intelligent stealth drivers and proxy rotation to bypass enterprise-grade WAFs and anti-bot protections.
  • Optimized data parsing algorithms to convert unstructured HTML into structured JSON/Markdown for Vector Database ingestion, ensuring data security.
  • Built an Agentic Crawl Orchestrator that uses an LLM agent to dynamically decide crawl depth, priority URLs, and content filters based on the target knowledge domain — replacing static crawl configs with goal-driven, adaptive ingestion.

CloudLens - Serverless Event Pipeline

PythonAWS LambdaTerraformAzure Functions
  • Designed an Event-Driven Architecture on AWS Lambda to automatically process and categorize high-volume media assets.
  • Implemented Infrastructure as Code (IaC) using Terraform to provision reproducible staging and production environments across AWS and Azure.
  • Integrated CloudWatch and Prometheus for full-stack observability and automated alert handling.

StreamGuard - Real-Time Fraud Analytics

PythonKafkaSpark StreamingPySpark
  • Architected a scalable streaming pipeline using Apache Kafka to ingest and process 10k+ transaction events per second.
  • Developed PySpark microservices deployed on Kubernetes to detect fraud anomalies with sub-200ms end-to-end latency, incorporating ML models.
  • Integrated an Autonomous Fraud Investigation Agent that, upon anomaly detection, autonomously queries transaction history, cross-references user behavior profiles, and generates a structured risk report — reducing analyst investigation time by 50%.

OmniPublish - Content Orchestration Engine

PythonSystem DesignMicroservicesNode.jsLangGraph
  • Architected a robust API Gateway using the Adapter Pattern to unify 17+ external platforms (LinkedIn, Twitter) into a single interface.
  • Implemented Circuit Breaker patterns and asynchronous retry mechanisms (Celery) to ensure fault tolerance across distributed APIs, with full-stack integration.
  • Embedded a LangGraph-powered Content Agent that autonomously adapts a single source post into platform-optimized variants (tone, length, hashtags, format) for each of the 17+ platforms using LLM tool calls, reducing manual content editing effort by 70%.

Technologies Across All Projects

TypeScriptReactAstroPythonCloudflare WorkersFirebaseRazorpayLangGraphLangChainAutoGenCrewAIOpenAINeo4jDockerKubernetesFastAPIPlaywrightGPT-4oTransformersONNX RuntimePyTorchEdge AILlama-3TensorFlow LiteRedisDistributed CrawlingSeleniumAWS LambdaTerraformAzure FunctionsKafkaSpark StreamingPySparkSystem DesignMicroservicesNode.js
36 unique technologies across 10 projects