Top Builders

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

LangChain

Large language models (LLMs) are emerging as a transformative technology, enabling developers to build applications that they previously could not. But using these LLMs in isolation is often not enough to create a truly powerful app - the real power comes when you are able to combine them with other sources of computation or knowledge. This library is aimed at assisting in the development of those types of applications.

General
Repositoryhttps://github.com/hwchase17/langchain
TypeLarge Language Model framework

LangChain - Resources

Resources to get stared with LangChain

LangChain - Use cases

Use cases for LangChain

LangChain - Example Projects

Implementations of LangChain

Edit on GitHub

Langchain AI Technologies Hackathon projects

Discover innovative solutions crafted with Langchain AI Technologies, developed by our community members during our engaging hackathons.

SignedOff — Verifiable Dependency Compliance

SignedOff — Verifiable Dependency Compliance

SignedOff is an autonomous compliance officer for Python software supply chains. Given a requirements.txt, it resolves the full dependency tree, evaluates every package's license against your declared use case, identifies known CVEs, generates contextualized risk scores, and routes findings through a policy-driven decision gate with human-in-the-loop review. What makes SignedOff different from existing scanners (Snyk, FOSSA, Black Duck, Dependabot): 1. Hash-chained, tamper-evident audit trail. Every decision — automatic or human — is sealed with SHA-256 into a verifiable chain. Tampering with any past entry invalidates all subsequent hashes. Built for SOC 2, FedRAMP, CMMC, and HIPAA audits where "show me your evidence" is the question that matters. 2. Use-case-aware risk contextualization. The same Django CVE has different urgency in a public-facing SaaS deployment versus internal tooling versus a distributed binary. SignedOff's LLM contextualizes severity against your declared deployment model — not generic CVSS scores. 3. Citation integrity guarantees. LLMs interpret evidence; they never generate citation URLs or identifiers. Every finding is backed by an OSV.dev record, GHSA advisory, or SPDX license entry — never hallucinated. The "citation integrity guardrail" forces human review when evidence is weak, regardless of severity. 4. Two-dimensional risk. License risk and security risk shown as parallel dimensions, never fused. Legal and security teams read the same dashboard and get their own answers. 5. Policy-as-code. POLICY.yml declares organizational risk tolerance in version-controlled YAML. Per-use-case license rules, severity thresholds, citation requirements. Applied uniformly across every reviewer. Built solo over 8 days with LangGraph, FastAPI, Anthropic API, OSV.dev. Open data sources only — no vendor lock-in. Try it: https://signedoff.onrender.com

VISION-LINK AI INNOVATORS

VISION-LINK AI INNOVATORS

VISION-LINK AI INNOVATORS presents a production-ready, automated multi-agent AI orchestration architecture built specifically to solve the high-cost hardware dependency barrier for enterprises and deep research domains like genomics. Core Architecture & Technical Approach Our system operates on a decentralized multi-agent ecosystem managed by a dynamic backend orchestration layer. Instead of relying on expensive, heavy, and power-consuming local GPU setups, our framework leverages distributed remote HuggingFace Inference APIs. The core orchestration intelligence dynamically loads, queries, and switches between specialized large language models and analytical models based on real-time task complexity. Dynamic Fitness & Schema Validation The backbone of our repository relies on custom runtime fitness metrics that continuously monitor API latency, response accuracy, and token efficiency. To ensure enterprise-grade data integrity, we implemented a strict live schema validation layer that screens all JSON data packets before they are parsed into production databases, ensuring zero structure mismatches during high-throughput operations. Strategic Impact By utilizing optimized remote compute endpoints rather than continuous idle local hardware, our architecture significantly lowers operational costs (OpEx) for scaling enterprise AI workflows. This makes advanced intelligent systems accessible, highly adaptable, and environmentally cost-effective for modern industries.

Maven — Cloud AI That Learns

Maven — Cloud AI That Learns

Most AI assistants feel disposable—you chat, you close the tab, and nothing sticks. Maven is different: a place where your AI agent can actually grow and learn inside a sandboxed cloud environment, without the heavy setup of hosting models locally or tuning everything yourself. Each Maven sandbox is your companion’s own space on Google Cloud Run: a persistent workspace filesystem, session memory, and tools you turn on as it matures—run Python scripts, build reports and charts, search documents, browse the web, read your Google Calendar, send Gmail, and more. You configure the personality and capabilities once; Maven handles deployment, isolation, and the hard calibration of running a capable agent in production. Because everything lives in the cloud, you are not tied to one laptop. Use the Electron dashboard to deploy sandboxes, upload docs, and watch files appear in the workspace browser—or message your agent from Google Chat on your phone and pick which sandbox to talk to. Switch between companions for work, research, or life admin; each sandbox stays separate with its own tools and growing file history. Maven is built for flexibility: one platform, many agents, each evolving in its own sandbox. Less “chatbot for a day,” more AI companion that accumulates context, output, and skill over time—wherever you are.

PakAI Capital | AI Investment Intelligence for PSX

PakAI Capital | AI Investment Intelligence for PSX

PakAI Capital is an enterprise-grade autonomous AI Investment Intelligence Platform that replaces traditional chartered financial analysts for Pakistan Stock Exchange (PSX) investors. Built with Google Gemini 2.0 Flash, the system deploys a coordinated team of six specialized agents: Research Analyst Agent — Performs fundamental analysis using RAG over company filings Macro Economist Agent — Evaluates SBP policy, PKR, inflation, and global signals Sentiment Analyst Agent — Processes real-time news from Dawn, Business Recorder, and more Risk Manager Agent — Calculates volatility, position sizing, and risk scores Portfolio Manager Agent — Synthesizes all signals into clear BUY/HOLD/SELL recommendations Governance & Compliance Agent — Audits every recommendation for hallucinations, bias, and missing disclosures Key Highlights: Real-time PSX data ingestion + news scraping ChromaDB / Pinecone RAG pipeline for grounded analysis LangGraph orchestration for reliable multi-agent workflow Beautiful Next.js dashboard with live agent reasoning feed Strong focus on explainability and AI governance — directly aligned with DeepMind’s enterprise AI priorities This submission targets Track 2: AI Agents with Google AI Studio and Track 1: Agent Security & AI Governance.

SentinelMesh — Enterprise AI Security

SentinelMesh — Enterprise AI Security

Enterprise AI agents are powerful — but ungoverned. A single manipulated prompt can leak credentials, exfiltrate PII, or trigger unauthorized actions inside production systems. Most teams have no visibility into what their agents are doing until something breaks. SentinelMesh solves this with a three-layer security architecture built on top of Lobster Trap, a custom Go-based Deep Prompt Inspection proxy: • Ingress DPI: Every outgoing prompt inspected in under 1ms via regex pattern matching. Prompt injections, PII requests, malware generation, data exfiltration, obfuscation, and sensitive path access blocked before they reach the model. • Egress DPI: Every model response buffered and inspected before delivery. Credentials and PII in outputs blocked at the proxy layer. • Intent Mismatch Detection: Each agent declares its intent per-request. Lobster Trap independently classifies actual intent. Mismatches trigger HUMAN_REVIEW — compliance officers approve or reject from the live dashboard. Three-layer RBAC is enforced independently at: the LangGraph orchestrator, the ChromaDB vector store, and the Lobster Trap policy file. Any single layer can block a request. The result: a 14-vector adversarial test suite with all attacks blocked, a live governance dashboard with real-time audit logs and attack heatmaps, and a Railway-deployed backend enterprise security teams can evaluate today.

Bridge-PA: Autonomous Prior Authorization System

Bridge-PA: Autonomous Prior Authorization System

The Problem: The healthcare Prior Authorization (Concurrent Review) process is notorious for being manual, slow, and prone to administrative bottlenecks, leading to delayed patient care and high costs for payer organizations. Our Solution - Bridge-PA: We built a multi-agent orchestration pipeline designed to securely automate this workflow while strictly adhering to HIPAA regulations. How it Works (The Prototype Architecture): Powered by LangGraph and LangChain, our deterministic state machine orchestrates four specialized AI agents. (Note: For this hackathon prototype, we are using simulated medical data and mock endpoints to demonstrate the workflow): 1.Supervisor Agent: Controls the fixed sequence and enforces system safety. 2.Document Processing Agent: Simulates the extraction of structured clinical data. 3.Criteria Evaluation Agent: Evaluates data against mocked healthcare policies for whitelist determination. 4.Data Entry Agent: Demonstrates how approved cases would be autonomously submitted. The system features smart routing: "Mode A" for end-to-end automation of clean cases, and "Mode B" for human-in-the-loop escalation where a Full Recommendation Package is sent to a UM Specialist. Note to Judges: Our complete LangGraph architecture works flawlessly in our local testing environment (as showcased in our demo video). During the final cloud deployment on Railway, we encountered a strict pip dependency conflict in the cloud environment. While the frontend UI is active, the cloud backend may experience startup limits. We highly encourage you to review our Demo Video, Pitch Deck, and GitHub repository to see the robust implementation of our orchestration pipeline!