Discover what the world's best startups are building before it's announced publicly.
Data Extraction
Model Evaluation
Prediction
Optimization
Fraud and Risk Scoring
Robotics teams delete 96% of their sensor data because they cannot move it fast enough. Byteport's DART protocol achieves 1500x faster transfer than TCP for large files, which turns a data bottleneck into a data asset for any team that generates more than it can ship.
RAG
Data Extraction
Computer Vision
Model Evaluation
Prediction
Data center construction is booming and electrical contractors are the bottleneck. Bidflow automates the estimating and bidding paperwork that keeps them from taking more jobs, starting with a vertical no general-purpose tool covers well.
RAG
Data Extraction
Speech AI
Model Evaluation
Optimization
Value-based care pays providers for outcomes, but most practices still chase quality gaps manually. Beacon Health automates gap closure inside the EHR with a live deployment covering 40,000 patients, which is real traction in a category full of pilots.
RAG
Data Extraction
Computer Vision
Speech AI
Model Evaluation
AI accounting startups like Zeni and Pilot still charge for human accountant hours. Balance pairs an always-on AI agent with human review only for filing, and a former CFO who built the bookkeeping himself is running it.
RAG
Data Extraction
Model Evaluation
Prediction
Optimization
Teams building AI agents hit a wall when general-purpose LLMs are too slow, too expensive, or not reliable enough for production. Burt helps them train specialized models that are 10x faster and cheaper. Both founders shipped production LLM systems at Replo (YC S21) and saw this gap firsthand.
Model Evaluation
Prediction
Optimization
Simulation
Fraud and Risk Scoring
Space power demand is expected to grow 500x by 2030 (orbital datacenters, stations, lunar outposts), and no one can launch enough solar arrays to meet it. Beyond Reach's deployable structures expand from a dining table to a football field in orbit, with $175M+ in LOIs already signed.
Data Extraction
Prediction
Optimization
Personalization
Anomaly Detection
Airlines and Uber proved dynamic pricing works for perishable inventory. Salons, fitness studios, and med-spas sell perishable time slots too, but price them statically. Booko applies the same math to a market that has never had it.
Data Extraction
Speech AI
Prediction
Personalization
Recommendation Systems
Hardware engineers waste hours cross-referencing datasheets across distributor sites. BaseFrame embeds a parts copilot directly in Slack and Teams, which is where the buying conversations already happen.
Data Extraction
Computer Vision
Model Evaluation
Prediction
Personalization
Zapier and Make require users to think in workflows. Bubble Lab's Pearl lives in Slack and runs multi-step operations (Jira, Stripe, Notion) from plain English, which means ops teams adopt it without learning a new tool.
Data Extraction
Speech AI
Model Evaluation
Simulation
Personalization
Instead of detecting fraud after the fact, BeeSafe deploys AI agents that engage scammers for months, mapping mule accounts and attacker infrastructure in real time. Three PhD founders literally wrote the first paper on this approach.
Data Extraction
Speech AI
Model Evaluation
Prediction
Personalization
Intent data incumbents like Bombora and 6sense sell stale, aggregated signals at six-figure minimums. Autumn scrapes real-time buying signals from GitHub commits, blog posts, and social media, then delivers them to Slack before the prospect has even started evaluating vendors.
RAG
Data Extraction
Speech AI
Model Evaluation
Personalization
Architecture firms spend more time on specs, compliance, and documentation than on design. Avoice already manages $300M+ in active projects across five countries, proving the workflow sticks in an industry that resists new software.
Computer Vision
Model Evaluation
Personalization
Recommendation Systems
Natural Language Processing
Chollet wrote the benchmark every frontier lab uses to measure AGI progress, and now he controls the next version. That gives ARC Prize a chokehold on how the industry defines and funds intelligence research.
Computer Vision
Speech AI
Model Evaluation
Simulation
Fraud and Risk Scoring
AI agents are shipping to production faster than anyone can test them. Ashr generates synthetic users that stress-test agents across text, voice, and images before real users hit the failure modes.
Data Extraction
Computer Vision
Model Evaluation
Prediction
Optimization
SAR satellites see through clouds and darkness, but the data is unusable without expert interpretation. AxionOrbital's foundation model converts raw SAR into optical-quality imagery at 1/100th the cost of tasking an optical satellite, making 24/7 earth observation accessible to any buyer.