Top Builders

Explore the top contributors showcasing the highest number of app submissions within our community.

LlamaIndex: a Data Framework for LLM Applications

LlamaIndex is an open source data framework that allows you to connect custom data sources to large language models (LLMs) like GPT-4, Claude, Cohere LLMs or AI21 Studio. It provides tools for ingesting, indexing, and querying data to build powerful AI applications augmented by your own knowledge.

General
AuthorLlamaIndex
Repositoryhttps://github.com/jerryjliu/llama_index
TypeData framework for LLM applications

Key Features of LlamaIndex

  • Data Ingestion: Easily connect to existing data sources like APIs, documents, databases, etc. and ingest data in various formats.
  • Data Indexing: Store and structure ingested data for optimized retrieval and usage with LLMs. Integrate with vector stores and databases.
  • Query Interface: LlamaIndex provides a simple prompt-based interface to query your indexed data. Ask a question in natural language and get an LLM-powered response augmented with your data.
  • Flexible & Customizable: LlamaIndex is designed to be highly flexible. You can customize data connectors, indices, retrieval, and other components to fit your use case.

How to Get Started with LlamaIndex

LlamaIndex is open source and available on GitHub. Visit the repo to install the Python package, access documentation, guides, examples, and join the community:

AI Tutorials

LlamaIndex Libraries

A curated list of libraries and technologies to help you build great projects with LlamaIndex.

Edit on GitHub

LlamaIndex AI technology page Hackathon projects

Discover innovative solutions crafted with LlamaIndex AI technology page, developed by our community members during our engaging hackathons.

Frugal Router: Cost-Aware Multi-Model AI Agent

Frugal Router: Cost-Aware Multi-Model AI Agent

Enterprises don't need a frontier model for every request. This agent detects which of the 8 task categories (factual, math reasoning, sentiment, summarization, NER, code debugging, logical/deductive reasoning, code generation) a prompt belongs to, then routes accordingly: code and constraint-satisfaction puzzles go straight to a frontier Fireworks model since small models are unreliable there and a failed cheap attempt costs tokens in addition to the retry; sentiment/NER/factual questions try a bundled 1.5B local model first (zero Fireworks tokens) and fall back only when it isn't confident. The difficulty classifier for math/factual ambiguity is trained on real data — MATH's Level 1-5 labels and RouteLLM's Chatbot Arena win/loss judgments — rather than hand-tuned guesses. Model tiers are built dynamically from ALLOWED_MODELS at container startup since the permitted model list isn't published until launch day.

Intelligent GitHub Repository Analyzer

Intelligent GitHub Repository Analyzer

CodeAtlas AI is an AI-powered GitHub Intelligence Platform designed to simplify the process of understanding, exploring, and contributing to open-source and enterprise software projects. Developers often spend hours navigating unfamiliar repositories, reading documentation, and tracing code before becoming productive. CodeAtlas AI dramatically reduces this onboarding time by automatically analyzing a GitHub repository and generating meaningful insights through artificial intelligence. The platform provides an interactive architecture overview, intelligent repository summaries, dependency visualization, repository health analysis, AI-generated documentation, and personalized learning paths for new contributors. Users can ask natural-language questions about the repository, receive explanations of important modules, discover where key functionality is implemented, and understand relationships between components without manually reading hundreds of files. Our solution combines GitHub APIs with modern large language models running on AMD-supported infrastructure to transform raw source code into actionable knowledge. The platform emphasizes usability, software engineering best practices, and developer productivity while leveraging AI to automate repetitive repository analysis tasks. By reducing onboarding complexity and improving repository comprehension, CodeAtlas AI benefits students, open-source contributors, software engineers, and development teams. Instead of spending hours understanding a new project, developers can gain architectural insights, identify critical files, assess project quality, and begin contributing confidently within minutes. Our vision is to build an intelligent developer copilot that makes every GitHub repository easier to understand, maintain, and evolve.

TokenEdge — Hybrid AI Routing Agent

TokenEdge — Hybrid AI Routing Agent

TokenEdge is a hybrid, token-efficient AI routing agent built for the AMD Developer Hackathon Token Golf challenge. Its goal is to complete a diverse set of AI tasks while using the fewest possible scored Fireworks AI tokens without falling below the required accuracy threshold. Instead of sending every task to a hosted language model, TokenEdge uses a verification-gated cascade. It first classifies and deduplicates incoming tasks locally. Deterministic solvers then handle suitable tasks such as mathematics, logical reasoning, named-entity recognition, formatting operations, and code execution at zero Fireworks token cost. When deterministic solving is not possible, a bundled local language model attempts the task. Its response is accepted only when confidence and structural verification checks pass. Tasks that remain uncertain are routed to the cheapest suitable model available through Fireworks AI. More powerful models are used only as a final accuracy backstop. The system includes category-specific output limits, minimal prompts, caching, model allow-list detection, automatic fallbacks, sandboxed code execution, and answer verification. Reasoning is disabled by default and enabled only when genuinely required. This architecture reduces unnecessary API usage while avoiding the zero-API-call disqualification condition. TokenEdge demonstrates how local computation, verification, dynamic model routing, and selective cloud inference can make AI agents significantly more efficient, reliable, and cost-effective.

Parallax

Parallax

PARALLAX is a GenAI-native cyber-fraud investigation platform for banks, fraud teams, SOC analysts, and cyber cells responding to APK-led financial crime. It analyzes suspicious Android applications and related fraud evidence to identify banking-malware behavior such as accessibility abuse, overlay attacks, SMS or notification interception, suspicious permissions, network indicators, and runtime compromise signals. PARALLAX converts these findings into structured evidence, confidence-scored claims, risk assessments, and investigation artifacts. The system is designed around an evidence-first workflow: ingest a suspicious APK, run static and dynamic analysis, extract IOCs and observations, snapshot the evidence bundle, coordinate specialist AI agents, surface challenges or disagreements, and produce a final human-reviewable action packet. Instead of treating fraud review as a static report, PARALLAX models it as a live investigation. Specialist agents can assess device compromise, transaction traces, mule-account patterns, evidence quality, liability context, legal evidence requirements, and final recommendations. The result is an auditable reasoning trail that helps teams understand not just what the system concluded, but why. PARALLAX keeps humans in control. Its outputs are designed to support analysts and bank officers with clear evidence, unresolved questions, recommended next steps, and exportable documentation for escalation or cyber-cell reporting.

SATRIA: AI Triage & Queue for Indonesia Healthcare

SATRIA: AI Triage & Queue for Indonesia Healthcare

SATRIA (AI-Powered Patient Triage and Queue Routing) is a digital transformation solution designed for Indonesian primary healthcare (Puskesmas and clinics). Utilizing a three-agent cooperative orchestration powered by the Band framework and AIML API (Claude-3.5-Sonnet & Gemini-2.5-Pro), SATRIA automates the intake and clinical evaluation pipeline: 1. Anamnesis Agent (@anamnesis): Guides patients in natural language (Indonesian/English) to collect demographic data (via NIK parsing) and clinical history. 2. Clinical Expert Agent (@clinical-expert): Leverages ChromaDB vector search (RAG) over official Kemenkes clinical guidelines (PNPK) to calculate patient triage priority (P1/P2/P3) and automatically book queues at the nearest facility. 3. Compliance Auditor Agent (@compliance-auditor): Conducts clinical safety checks (drug interactions, allergies) and compiles structured SOAP Electronic Medical Records (EMR) according to PMK No. 24/2022. The system connects an Astro/Tailwind frontend, a high-concurrency FastAPI gateway, and Redis caching. It reduces patient queue times, prevents clinical errors, and relieves administrative burdens from doctors.

InvestiGuard AI

InvestiGuard AI

InvestiGuard AI is a multi-agent fraud investigation platform built for regulated and high-stakes financial workflows. Traditional fraud detection systems often provide risk scores with limited transparency, making investigations difficult to audit and explain. My solution uses Band to coordinate five specialized AI agents that collaborate throughout an investigation. The Intake Agent triages suspicious cases, the Transaction Analyst examines transaction patterns, the Behaviour Analyst reviews customer activity, the Risk Assessor evaluates evidence and calculates risk, and the Lead Investigator generates the final recommendation. Rather than relying on a single AI system, agents exchange messages, delegate tasks, share findings, and build upon each other’s analysis through a structured collaboration workflow. The platform visualizes this process through an Agent Collaboration Network, Band activity feed, evidence board, investigation timeline, and collaboration metrics such as handoffs, messages exchanged, active agents, and confidence scores. To support accountability, InvestiGuard AI includes a human-in-the-loop review stage where investigators can inspect evidence, review recommendations, approve or override decisions, and generate a complete audit report. The demo showcases multiple fraud scenarios including transaction structuring, suspicious international wire transfers, and legitimate high-value transactions. By combining multi-agent collaboration, explainable reasoning, human oversight, and auditable decision-making, InvestiGuard AI demonstrates how Band can power trustworthy AI systems for fraud detection, risk assessment, and financial compliance workflows.