Show HN: Tako, a Knowledge Search API

[ttobbaybbob]

I'm Bobby, CTO of Tako. We just launched our Knowledge Search API.

Our API takes natural-language prompts like "Nvidia M&A history" and returns visual answers and grounding text sourced from real-time, structured data (example: https://trytako.com/card/YHloo1Ea7GRnBr_s5r6s/).

Most AI systems can’t effectively reason about real-time, structured data. One reason is access: a lot of the most valuable info is trapped in databases web crawlers can't index. Google solves this with a team of 2k+ engineers that ingest data (stocks, sports, etc) into a proprietary Knowledge Graph. Our goal is to offer developers the same knowledge search + visualization primitives Google has built, tailored for AI use cases, and delivered via API.

We seek to augment LLM’s capabilities, and this means that most of our biggest technical challenges stem from not getting a lot for “free” from LLMs. For example, RAG architectures that generate final outputs with LLMs introduce accuracy issues we can’t tolerate, and are too slow. We’ve built a Generative Augmented Search (GAS) architecture that uses LLMs (currently Llama 3.3-70B on Cerebras) to analyze input queries (~200 ms) but use deterministic retrieval for most output generation. The data in our knowledge graph generally isn’t available in LLMs or the web, so we have to acquire it directly from sources (including licensing it from authoritative providers like S&P Global). A limitation of this approach is that some developers want us to offer the flexibility of LLM analysis across web sources, even if it means tolerating non-authoritative sourcing and some hallucination. We’re working on some solutions to that now.

Curious to hear how other people are fighting hallucination.

I'd love your feedback on the product (and happy to discuss/answer questions about it/the tech stack)

[IntToDouble]

The first rule of Show HN is, "don't give me a playground where it looks like I can push the button and then force me to auth when I actually do click the button, expecting magic on the other side."

The second rule of Show HN is, "YOU DO NOT FORCE USERS TO AUTHENTICATE AFTER THEY'VE CLICKED THE BUTTON."

[ttobbaybbob]

:saluting_face we debated whether to do or not to do it like this. We were scared of people scraping us (among other things) and decided to be bad

[nicholashandel]

What kinds of things are you seeing people build? Curious on the use cases in prod!

[ttobbaybbob]

Overall, we improve any AI answer - we've been integrated into AI search experiences (the most common/obvious use case), content generation use cases (eg https://capitol.ai/), but we're excited to see what else people come up with!

[seongboii]

Super interesting. Love the flexibility on this

[ttobbaybbob]

Thanks for trying it out!

[jake-jung]

How is GAS different from RAG?

[dragonwriter]

The description seems to invert RAG: RAG (simplified a bit) conducts a search, presents the results to an LLM, and then the LLM produces output based on the search. In GAS as described the LLM analyzes the input query to generate the search query, then a search is run, and the results of the search are presented to the user. With GAS, you always get specific authoritative (to the extent that describes the universe searched) documents, any hallucination will impact document selection not correctness of info. Wtih RAG you get an LLM analysis of potentially many authoritative (with the same caveats as before) documents, but hallucination can affect the accuracy of the presented response as well as its relevance.

[huckfinn]

Does this work well with MCP?

[ttobbaybbob]

It does (and we're working on making it better)! Check out our docs here -> https://docs.trytako.com/documentation/integrations-and-exam...