Valyu Wins Hebbia’s TECH WEEK Pitch

Late Tuesday afternoon on the 3rd of June, Hebbia’s SoHo loft had been reshaped into an intimate demo theatre for TECH WEEK by a16z. Chairs faced a compact stage; spotlights spilled across laptops, cables, and half-finished coffee cups. Five teams would give 5-minute pitches, but the friendship-and-banter vibe hid a quiet undercurrent of nerves.

Our co-founder, Alexander Ng, rehearses in the corner. He had trimmed his fifteen-slide deck down to a 5-minute sprint, murmuring his opener under his breath. Alex isn’t usually one for theatrics, yet there he was, pausing on each slide to make sure the transitions felt natural. Nervous, yes, he later admitted, but also determined to show what DeepSearch can do.

The Problem We Walked On Stage to Solve

Modern language models seem to read the open web, but anyone who has built a real-world agent knows the truth:

• Signal-to-noise is low. SEO is louder than substance, leaving shallow snippets where research should be.
• Important sources are locked away. Academic and market data sit behind paywalls or PDFs that don’t scrape cleanly.
• Glue code takes over. Dev hours that should go to product creep into retry logic, schema fixes, and brittle scrapers.

Our short pitch asked the room to imagine a different default: LLMs that have access to the content that they need, trustworthy metadata, and citations that link back to primary research.

The 180-Second Solution

Alex told that story in three short beats.

• High-value sources. We start by feeding Valyu only top-shelf material—PubMed papers, every arXiv pre-print, and (in a pilot) Wiley’s entire Business, Finance & Accounting catalogue—so the model never has to rely on the open web for credibility.
• LLM-ready structure. That content is parsed, indexed at a petabyte scale, and tagged with clean, section-level metadata, so an agent can reason over full documents, not fragments or broken tables.
• One-call delivery. A single DeepSearch endpoint turns any prompt into citation-rich JSON in under a second, meaning developers skip the glue code and agents get answers they can trust.

What the Judges Noticed

The feedback came back plain and direct, which we appreciated:

1. Works today. “We could plug this in tomorrow.”
2. Fits the model ecosystem. “Higher-grade inputs make every provider better.”
3. Grows with usage. “Each agent query becomes its own revenue event.”

Hearing our own priorities reflected back to us was a welcome signal that the problem and the solution both resonate beyond our team.

A Week of Momentum

Since the pitch night:

• Builders have spun up API keys and dropped DeepSearch into their AI stack.
• Publishers have reached out about licensing their corpora.

We’ve been answering those emails between pull requests, grateful and a little humbled by the attention.

Next up, we’re scaling pilots, loading the next domain corpus, and updating the public playground. Thanks to Hebbia, the judges, and every builder who poked holes in the demo—your questions sharpen the product.

Quiet today; tomorrow we’re back to cranking indexes and shipping code.

Try out the Deep Search API.

👉🏻 >1000 free queries

👉🏻 >Fully cited responses to ground your agents and LLMs

👉🏻 >Slot it easily into your workflow with 4 lines of code or our API

1from valyu import Valyu
2
3valyu = Valyu(api_key="your-valyu-api-key")
4
5response = valyu.search(
6  "Extending context window of large language models via positional interpolation",
7  search_type="proprietary",
8  max_num_results=5, 
9  max_price=30, 
10  included_sources=["valyu/valyu-arxiv"]
11)
12
13# Get detailed academic content with citations
14print (response)
Copy