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.

ABI (Automated Business Intelligence)

ABI (Automated Business Intelligence)

Corporations are drowning in data but starving for immediate insight. Traditional Business Intelligence (BI) workflows require data engineering pipelines, complex SQL dashboards, and manual assembly of PowerPoint presentations just to share weekly updates. Business leaders lose critical windows of opportunity waiting for static reports that fail to offer interactive depth. Abi is an automated enterprise BI assistant that bridges the gap between raw corporate data and presentation-ready insights. Users simply connect their data sources and organize them into workspace folders. Abi instantly takes over, allowing users to explore their data through intuitive conversational chat or automatically spin up comprehensive, publication-grade analytical reports. The core architectural engine is built using LangGraph, establishing a modular, robust multi-agent pipeline. It features: Dynamic Intent Classification: An intelligent routing system that determines whether a user needs a casual data answer or a deep-dive report pipeline. Autonomous Sandbox Exploration: A secure Python REPL environment where the agent writes, tests, and executes Pandas code directly against datasets (like our functional supermarketsales.csv deployment) to pull real-time calculations without risking data exposure. What We Built & The Future Roadmap Due to the hackathon timeline constraints, our current submission showcases the critical validation engine: a fully functional conversation analyst node that dynamically ingests, queries, and interprets complex retail transactions. The groundwork is explicitly laid for Abi's definitive killer feature: transforming structured dashboard data into fully automated document reports (.pptx, .pdf, .xlsx) and automatically synthesizing them into a synthetic video presentation. Abi turns static numbers into a shareable, human-like data briefing video with a single click. Abi is the future of hands-free corporate reporting.

Smart Selector

Smart Selector

Leveraged the Fireworks AI and its blog on best models to choose from. This project helps you minimize operational costs by intelligently selecting the most appropriate model for each task. Instead of always defaulting to large, expensive models, the system evaluates the complexity of the input and chooses a model that balances accuracy, speed, and token usage. By reducing unnecessary token consumption, you not only save on API costs but also improve overall performance. For example: Short queries → lightweight models are chosen to keep token usage minimal. Complex tasks → larger models are selected only when needed, ensuring accuracy without waste. Adaptive learning → the selector continuously refines its choices based on prior results, making future runs even more cost‑efficient.

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.

Clip Insighs

Clip Insighs

Clip Insights is a Chrome extension that transforms the way you learn from videos by turning hours of footage into fast, actionable notes and insights. Instead of scrubbing through long videos to find what matters, users can take screenshots, make notes, get rapid AI-generated summaries, along with key points extracted from any video. The extension supports interactive video chat, letting viewers ask questions directly about the content and receive instant, real-time answers. Users can also capture instant screenshots at any moment in a video and take timestamped, unlimited notes right inside the extension. Everything - summaries, screenshots, and notes - can be compiled and exported into a single downloadable PDF with one click, making it easy to save, organize, and revisit learning material. Built for students, researchers, and lifelong learners, Clip Insights streamlines video-based learning into one seamless, distraction-free workflow.

HealthOS — Turn Lab Reports Into Real Answers

HealthOS — Turn Lab Reports Into Real Answers

The average person typically receives a bunch of numbers in their lab report with no idea what is normal, no understanding of improvement, and no quick method to query their lab report — without having to search the web at 2 am and risking a site that’s not reputable. HealthOS fixes that. Simply upload a lab report (PDF file or even a picture) and HealthOS automatically pulls out all the values, highlights anything outside a healthy range and collapses it into a running health score that will update with each upload as it gets older. Each report follows on from the previous report and rather than just a single disconnected ‘snapshot' of progress, you are able to get an actual ‘snapshot' of improvement or not. It's accompanied by an AI chat assistant that already knows about your profile, medications, and lab history, allowing you to pose real questions to your bot and receive answers based on your own data, rather than general search data. HealthOS is a full stack application: OCR-powered report parsing, FastAPI backend, Structured health entries tracked over time in a PostgreSQL database, JWT-authenticated user accounts, with a Fireworks AI-powered assistant, and a React frontend, deployed end-to-end on free infrastructure (Vercel, Railway, Neon). We created HealthOS because the way we see it, knowing your health shouldn't require a medical degree or a doctor on speed dial — it should be something your data does for you.

DevMemAgent: Dev Memory on AMD ROCm

DevMemAgent: Dev Memory on AMD ROCm

DevMemAgent is a developer-productivity memory agent that gives coding assistants persistent, structured long-term memory across conversations. Today's LLM tools are amnesiac — every new chat starts from scratch, so preferences get re-stated, recurring build errors get re-debugged, and sprint context evaporates between threads. DevMemAgent fixes this by listening to developer sessions and persisting what matters: typed memories tagged as preference, fact, event, goal, or relationship. Built as a fork of the LangChain memory-agent template (MIT, attributed), it adds three differentiators over the upstream ReAct loop. First, a typed memory schema with five categories so retrieval can filter by what the user is actually asking about. Second, decay and confidence scoring — each memory carries a confidence score and optional TTL, with a recency-weighted effective score used to re-rank retrieved memories so stale or low-confidence facts naturally fade. Third, a memory graph with typed edges (depends_on, supersedes, contradicts, related_to, refines) enabling graph-aware recall for questions like "what changed since X?" The stack runs local-first on AMD ROCm / Instinct GPUs via Ollama — default models are llama3.1:8b for chat and bge-small-en for embeddings, both executing on-device. The Docker image uses rocm/pytorch:6.3.2-ubuntu-22.04-py311 with GPU passthrough through /dev/kfd and /dev/dri. Cloud providers (OpenAI, Anthropic) are optional fallbacks for A/B quality comparison but are never required. A CPU-only docker-compose override enables demos on any host. The submission includes 42 source files, 36 passing unit and integration tests, a 5-case recall/precision eval harness, a reproducible 3-minute demo script, a 12-slide pitch deck, and a Docker Compose stack that launches the agent with one command. The codebase includes a professional README, architecture documentation, and a NOTICE file crediting LangChain upstream. License: MIT (preserved from upstream).