Top Builders

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

Chroma

Chroma is building the database that learns. It is an open-source AI-native embedding database. Chroma makes it easy to build LLM apps by making knowledge, facts, and skills pluggable for LLMs. The fastest way to build Python or JavaScript LLM apps with memory

General
Relese date2023
AuthorChroma
Typeembedding database

Tutorials

Great tutorials on how to build with Chroma

Chroma - Helpful Resources

Check it out to become Chroma Master!

Chroma - Clients

Connect Chroma to your Project!

Chroma - integration

Get started with Chroma!

Edit on GitHub

Chroma AI technology page Hackathon projects

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

Dementia ASR Screening

Dementia ASR Screening

Overview The Dementia ASR Screening & Multi-Factorial Risk Stratification Application is an advanced digital health portal designed to bridge the gap between AI-driven speech biomarkers and modifiable clinical risk factors. Built on a premium, mobile-first dark theme interface, the portal enables patients to undergo non-invasive, visual speech elicitation tests in seconds, generating comprehensive, weighted dementia conversion risk assessments. Key Features & Technological Innovation Lossless Acoustic Elicitation: Features a custom-built, client-side Web Audio API recorder that encodes raw microphone samples directly into lossless 16-bit Mono PCM WAV format. This ensures 100% processing compatibility with cloud-based Automatic Speech Recognition engines. Speechmatics SaaS ASR Integration: Orchestrates advanced speech analyses via Express backend proxies. By disabling disfluency filtering (remove_disfluencies: false), the portal natively captures and tags spoken filler words. Acoustic Pause Biomarker Extraction: Reads word-level transcript timestamps to automatically isolate and flag long conversational hesitation gaps exceeding 1.5 seconds, a key marker in acoustic cognitive screens. Wiley 2025 Multi-Factorial Stratification: Incorporates a validated clinical risk algorithm based on the October 2025 study in Alzheimer's & Dementia (DOI: 10.1002/alz.70870): Age Scoring, Cognitive Reserve, BMI,Depression Screening (PHQ-2), Weighted Clinical Risk Dial: Automatically weighs Vocal speech biomarkers (40% weight) against Clinical modifiable factors (60% weight) to render a composite, color-coded Stratified Risk Indicator (Low, Moderate, or High Risk). Comprehensive Session Review Timeline: Connects to a self-healing SQLite database to store serialized disfluency details, allowing patients to inspect past tests and review historical reports exactly as they looked live along with an Interactive Demographics Editor.

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.

Onus HybridMind: AI Procurement Auditor

Onus HybridMind: AI Procurement Auditor

Enterprise procurement teams are bound by complex vendor contracts containing rebate thresholds, volume discounts, and penalty clauses buried in dense legal documents. Meanwhile, SQL databases hold the transaction records, but no one cross-references them against contract terms at scale. The result is millions in uncollected rebates and compliance gaps going undetected. To solve this, we built HybridMind: an autonomous AI audit agent that bridges structured SQL procurement data and unstructured vendor contracts in real time. It is not a standard RAG chatbot; it is a deterministic, multi-agent auditor built specifically to stop financial leakage. Our backend leverages LlamaIndex Workflows to orchestrate three distinct AI agents powered by Gemini 3.1 Flash Lite. The process begins with the Executor Agent. It converts natural language into strict SQL queries against our Supabase PostgreSQL database while simultaneously querying our ChromaDB vector store for the exact legal clauses. Next, the Verifier Agent takes both data sources and performs logical validation. It cross-references the SQL math against the PDF rules to detect discrepancies, such as verifying if a 12,000 unit purchase correctly triggers a 10,000 unit rebate threshold. Finally, the Chronicler Agent packages the verified finding and broadcasts the financial metrics to our live React dashboard via WebSockets. This architecture ensures transparent reasoning. While standard AI often hallucinates numbers, HybridMind forces the AI to show its raw SQL, explicitly cite its contract clauses, and prove its math. By keeping the frontend stateless and the backend firmly grounded in dual-silo retrieval, nothing is hidden behind a black box. Ultimately, HybridMind turns silent leakage into documented liability.

Benchmarking Robustness in Agentic RAG Systems

Benchmarking Robustness in Agentic RAG Systems

As AI agents become increasingly integrated into real-world applications, understanding retrieval reliability and preprocessing sensitivity has become a major challenge in Retrieval-Augmented Generation (RAG) systems. Most traditional evaluations focus only on architecture performance while ignoring how preprocessing decisions can significantly affect retrieval robustness and benchmark outcomes. In this project, we built an interactive observability and benchmarking platform for evaluating robustness in Agentic RAG systems. The platform compares Single-Agent and Multi-Agent RAG architectures across SQuAD and HotpotQA benchmarks using Exact Match (EM) and F1 evaluation metrics. Through systematic experiments, we discovered a key insight: preprocessing strategies such as chunking can completely flip benchmark winners. Without chunking, the Single-Agent system slightly outperformed the Multi-Agent system on SQuAD. However, after introducing chunking, the Multi-Agent architecture became significantly more robust under noisy retrieval conditions. To make these behaviors observable, we developed an interactive Streamlit dashboard featuring benchmark comparison analytics, retrieval trace visualization, chunking impact analysis, and failure inspection. One of the core components of the platform is the Retrieval Trace Viewer, which allows users to inspect how Multi-Agent systems rewrite queries, retrieve semantically richer chunks, and improve answer generation step-by-step. We also analyzed common RAG failure modes such as vocabulary mismatch, retrieval pollution, and chunk fragmentation. Our findings demonstrate that retrieval robustness depends not only on architecture design but also heavily on preprocessing strategy and retrieval quality. Technologies used include LangChain, LangGraph, FAISS, HuggingFace Embeddings, Groq LLMs, Streamlit, Plotly, and Python.

FinSight Agent — Crypto Intelligence Platform

FinSight Agent — Crypto Intelligence Platform

FinSight Agent is an AI-powered crypto intelligence platform for enterprise treasury management and portfolio monitoring in Southeast Asia. The Problem: Crypto holders lack affordable tools to monitor portfolio risk in real time. Manual monitoring misses threats that happen within minutes, and anomaly detection traditionally requires dedicated analysts. What It Does: Wallet Anomaly Detection — Monitors on-chain activity using Z-score and IQR statistical analysis. Detects six anomaly types: Rapid Drain, Large Transfer, Dust Attack, Frequency Spike, Failed Transaction Spike, and New Address Interaction. Each alert includes an AI-generated explanation. AI Market Signals — Combines RSI, MACD, Bollinger Bands, MA crossover, momentum, and Fear & Greed Index across 60+ coins. Signals explained by Groq/Llama 3.1 in natural language. Real-time Dashboard — Interactive web dashboard with live prices, switchable coin charts, market intelligence table, anomaly alerts, portfolio tracker, and AI chat interface. Telegram Bot — 15+ commands including /analisis, /scan, /status, /berita, /okx. Auto-scan runs every 30 minutes and delivers proactive alerts. OKX Integration — Connects to OKX CEX and DEX via authenticated API for live balance and portfolio data. Tech Stack: LangChain, Groq (Llama 3.1), Python, NumPy, Flask, SQLite, Etherscan API, OKX API, CoinGecko API, Telegram Bot API, Web3.py, Chart.js. Why Track 4: This system addresses all Track 4 criteria — multi-source data ingestion, AI-powered analytics, anomaly detection, and natural-language querying over stored intelligence. It transforms fragmented data from seven sources into validated, actionable enterprise intelligence. Target Users: SMEs in Southeast Asia managing crypto treasury, individual traders needing institutional-grade risk monitoring, and Web3 projects requiring wallet security without a dedicated security team.