Discover what the world's best startups are building before it's announced publicly.
Data Extraction
Speech AI
Model Evaluation
Prediction
Optimization
PE firms and search funds hire junior analysts to cold-email business owners about acquisitions. Q2Q automates the entire pipeline from thesis to scheduled meeting, with a 5,000+ company database. The founder has PE deal experience from Jefferies and the YC partner is Harshita Arora.
RAG
Data Extraction
Computer Vision
Model Evaluation
Optimization
Real-time threat detection in live surveillance for law enforcement. Pilot deployments with Atlanta, Chicago, and St. Louis police departments. The founder deployed reinforcement learning for video intelligence in classified defense environments at Lockheed Martin before starting Protent.
RAG
Data Extraction
Computer Vision
Speech AI
Model Evaluation
Early-stage founders handle customer success themselves because they cannot afford a CS team. Pollen's AI agents automate tickets, proactive outreach, and customer inquiries, giving solo founders the responsiveness of a dedicated team. Pivoted from a $4M-funded freelancer network after seeing the AI opportunity.
RAG
Data Extraction
Speech AI
Model Evaluation
Prediction
Teladoc and Amwell are reactive: you feel sick, you call. Prana monitors wearable data and medical records continuously, catching rising blood pressure or glucose trends before symptoms appear. The founder deferred Brown Medical School to build it, and the beta is live with $49 physician visits.
RAG
Data Extraction
Computer Vision
Speech AI
Model Evaluation
Traditional recruiting agencies charge 20-25% of first-year salary. Perfectly's AI agent sources, screens, and delivers candidates via Slack with 2x higher interview pass rates in 2-4 weeks. The founding team built TikTok's and Meta's recommender systems, so candidate matching is the same problem they already solved at billion-user scale.
Data Extraction
Computer Vision
Model Evaluation
Prediction
Anomaly Detection
AI coding agents execute well but plan poorly on complex features. OpenSpec adds a spec layer that structures the plan before any agent writes code, orchestrating 30+ agents through documented, reviewable workflows. The missing discipline layer for agentic development.
Data Extraction
Computer Vision
Speech AI
Model Evaluation
Prediction
Commercial insurance for high-risk industries (explosive manufacturers, hazmat trucking, steel mills) takes 7 days per policy through manual portal logins and underwriter negotiations. Panta's agents do it in 4 hours. Both founders are licensed brokers who have personally placed hundreds of complex policies.
RAG
Data Extraction
Model Evaluation
Optimization
Simulation
Grand strategy games (EU4, Hearts of Iron) have passionate audiences but static AI. Pax Historia replaces scripted opponents with LLM agents that negotiate, betray, and adapt, running across 30+ model providers with automatic failover. The game plays differently every time.
Data Extraction
Model Evaluation
Prediction
Personalization
Fraud and Risk Scoring
Crypto compliance is a patchwork of state-by-state licenses, AML/KYC rules, and transaction monitoring requirements. Most crypto startups cobble it together manually. Payna consolidates the full stack into one platform, built by Blockchain at Berkeley founders with hands-on DeFi experience.
Prediction
Optimization
Anomaly Detection
Reinforcement Learning
Natural Language Processing
GPU inference is bottlenecked by electrical data movement. Piris replaces it with photonic interconnects claiming 5x lower latency and 10x lower power per bit. The CEO holds multiple MIT degrees in EE and physics and built optical systems for NASA. High-risk, high-reward bet on a physical law advantage.
Data Extraction
Speech AI
Model Evaluation
Prediction
Optimization
Dental clinics miss calls, lose patients, and bleed revenue at the front desk. Patientdesk's AI receptionist books appointments, verifies insurance in real time during the call, and collects payments. One clinic generated $350K+ in revenue from AI-managed bookings in a single month. 60+ clinics across three countries.
Data Extraction
Computer Vision
Speech AI
Model Evaluation
Optimization
SOX internal audit at pre-IPO and public companies requires weeks of manual scoping, flowcharting, and evidence review. Oxus turns it into minutes with AI that reads spreadsheets, PDFs, and screenshots. Top-5 most upvoted launch in YC W26, built by a founder who did digital payments research at the Federal Reserve.
Data Extraction
Computer Vision
Speech AI
Model Evaluation
Prediction
Enterprises have no idea how many AI tools their employees use or what data flows through them. Oximy discovers every AI application from the network layer (no endpoint agents), classifies data in real time, and gives CISOs a single dashboard. SOC 2 Type I certified with an open-source sensor.
Data Extraction
Model Evaluation
Prediction
Optimization
Simulation
Startups park cash in checking accounts earning near-zero while bonds yield 4.5-5%. Palus connects via Plaid and moves idle cash into institutional bond portfolios with 1-2 day liquidity. No bank switch required. Built by someone who ran investment research for a $3T AUM asset manager.
Data Extraction
Computer Vision
Speech AI
Prediction
Personalization
Screen-free voice device with 120+ language transcription, speaker identification, and dynamic mind mapping. HIPAA and SOC 2 certified. The founder built and open-sourced Omi, one of the first AI note-taking hardware devices, so the hardware-software integration is not new territory for him.