> ## Documentation Index
> Fetch the complete documentation index at: https://guide.kello.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Get better results from Kello Alpha

> Ten habits from real Kello Alpha use: name the kind of company, say who to leave out, turn soft words into facts, narrow in follow-ups, ask why.

Kello Alpha gives its best results when you name the kind of company as well as the title, say who to leave out, and turn soft words like "recently" into dates. Start broad, narrow in short follow-ups, read the last sentence of every answer, and when a number looks wrong, ask Alpha why.

Every example below is a real ask, with what came back.

At a glance:

* [Read the last sentence](#read-the-last-sentence)
* [Name the kind of company, not just the title](#name-the-kind-of-company-not-just-the-title)
* [Say who to leave out](#say-who-to-leave-out)
* [Turn soft words into facts](#turn-soft-words-into-facts)
* [Start broad, then narrow](#start-broad-then-narrow)
* [If a number looks wrong, ask why](#if-a-number-looks-wrong-ask-why)
* [Ask for one list per message](#ask-for-one-list-per-message)
* [For one person, give a name and company, or a LinkedIn URL](#for-one-person-give-a-name-and-company-or-a-linkedin-url)
* [Paste example profiles in the same message](#paste-example-profiles-in-the-same-message)
* [For a team map, ask for everyone](#for-a-team-map-ask-for-everyone)
* [Ask for a number when you want a shortlist](#ask-for-a-number-when-you-want-a-shortlist)

## How does a search run?

Knowing the steps shows you where your words matter most.

**1. Writes a portrait**

Alpha writes a short portrait of who you want: who you are hunting, and who you are not. Every later step is checked against this portrait, so your words shape it more than anything else.

**2. Looks up names**

It looks up the names you gave: companies, schools and industries. Exact names help.

**3. Searches**

It searches across 830M+ people, using the filters it built from your ask.

**4. Reviews top results**

A second review checks the top results against the person you described. It looks for people who don't fit and for good people who were missed.

**5. Adjusts, at most twice**

It tightens or loosens, then checks again. At most twice.

**6. Saves the list**

It saves the list and shows it to you, with a list card you can open.

**7. Tells you what it did**

One or two sentences on what it assumed, widened or couldn't filter.

If a word in your ask is vague, the portrait guesses, and the whole run follows that guess. Most of the habits below are ways to stop it guessing.

## Read the last sentence

The last sentence or two of every answer is where Alpha tells you what it assumed, what it widened and what it couldn't filter.

> [!TIP]
>
> Read it before you open the list.

**Real example: a search with a salary band in it.**

> "Current compensation cannot be filtered directly from profiles, but targeting 10+ years of experience across AVP, Director, and VP-level transaction leads benchmarks closely to the 60–70 LPA (Lakhs per annum: yearly pay in Indian rupees (1 lakh = 100,000).) bracket."

People searches also stay in your home country unless you say otherwise. Alpha says so here too ("Searched in India"). Reply `search globally` or name the places you want.

If the last sentence shows a wrong guess, correct it in one line. You don't need to start again.

## Name the kind of company, not just the title

Titles mean different things in different places. "Principal" at a VC fund and "Principal" at a consulting firm are different jobs. Alpha does best when you say where the person works, and which jobs count.

**Real example: a deeptech fund.** A deeptech fund recruiter first got investor relations staff mixed in with investors. One follow-up fixed it:

> "I only want investing profiles … you have shown me investor relations"

Alpha dropped investor relations, fundraising, compliance, operations and finance staff, and kept 19 deal-makers.

```text Write it like this wrap theme={null}
VPs and Principals at deeptech or defence-focused VC funds in India, investing roles only.
```

## Say who to leave out

Alpha excludes what it knows to exclude. The people you'd reject at a glance are usually the ones to name.

| Ask | What leaked in until it was named |
| - | - |
| Recently joined engineers at Pocket FM, an audio company | Sound engineers and audio QC staff |
| Investors at deeptech funds | Investor relations and fundraising staff |
| Investing team at a PE firm | Fund admin, accounting, legal |
| AI builders in a city | IT-services body shops |
| Lawyers with IP and data-protection experience | General statutory compliance staff |

```text Write it like this wrap theme={null}
… exclude investor relations, fund operations and interns.
```

## Turn soft words into facts

Words like *recently*, *unicorn*, *early stage*, *senior* and *top-tier* each need a definition. If you don't give one, Alpha picks one and tells you in the last sentence.

* **"Recently joined the engineering team at Pocket FM"** gave 60 people. Alpha read "recently" as "since January 2025", which left out 95 engineers who joined in 2024. `Joined since Jan 2024` settles it.
* **"Folks in India who have worked on RL environments"** brought back robotics simulation people. The recruiter meant LLM training environments. Alpha checks the web when a term may have a new meaning, but one line still saves a run: `RL environments in the LLM sense: task worlds with verifiers and rewards for training models.`
* **"Quit OpenAI in the last 7 days"** can't be answered to the day. Job dates are stored by month, so Alpha returned the most recent movers and said why.

```text Write it like this wrap theme={null}
Left a company valued at $1B+ between Oct 2025 and now, and now hold a Founder or Co-founder title at a company founded in 2025 or later.
```

## Start broad, then narrow

Follow-ups reuse what Alpha already learned in the conversation. A first search takes about 3 minutes. A follow-up takes about 1. Several of the best lists came from three to five short follow-ups.

A real event guest list:

| Message | Result | Time |
| - | - | - |
| "find me right people to invite to this event" + lu.ma link | 10,000+ AI researchers in Bangalore | 51 s |
| "limit to those actively working in research right now" | Still 10,000+, now current research roles only | 58 s |
| "limit it to people who have PhDs" | 4,075 | 53 s |
| "limit to those currently based in bangalore and working on topics that will be covered in this edition of the event" | 1,087, ranked on the event's topics | 35 s |

And an FP\&A search: "FP\&A professionals with 5-8 years of experience based in Mumbai, good if they are in a listed company or one about to list" gave 376. "Out of the above list show me candidates who are working in a real estate company" gave 14, at developers and REITs.

Counts above 10,000 show as "10,000+", and the review reads the top of the list. If the top of a huge pool looks off, narrow it before you judge the search.

A clear first ask, then short follow-ups:

```text Write it like this wrap theme={null}
only India-based
```

```text Write it like this wrap theme={null}
limit only principals, no founders or partners
```

```text Write it like this wrap theme={null}
add professionals from [two named funds]
```

## If a number looks wrong, ask why

Alpha can explain any count it gives you. It's the fastest way to find the assumption that went wrong.

```text Write it like this wrap theme={null}
This seems low. What did you leave out?
```

**Real example: the Pocket FM search.**

> "This seems very low, what did you skip"

44 seconds later, on the Pocket FM search:

> "The count of 60 was driven by three specific boundaries: 1. Start date cutoff (January 2025+ vs. 2024+) … 95 joined in 2024 … 2. Audio & Creative 'Engineers' … \~150+ … were filtered out. 3. Interns & Contractors … \~40+ profiles were excluded … If you consider 'recently joined' to include 2024 hires … let me know and I will widen the list."

You now know exactly which line to change. The same works when a number looks too high, or when one person looks out of place: `why are these people on the list?`

## Ask for one list per message

Alpha shows one list per answer. Ask for three at once and you get the first.

**Real example: three lists of heads of legal.** One recruiter asked for heads of legal at startups that went public in the last 8 years, at startups that filed for IPO in the last 3 months, and at startups that haven't listed but whose legal head worked at one that did, "in separate lists". Alpha built the first (99 people), said the second couldn't be filtered because filing dates aren't in the data, and offered the third as the next step. "Now the third list" got it in 21 seconds.

One list now, then in the next message:

```text Write it like this wrap theme={null}
now the [second] list
```

## For one person, give a name and company, or a LinkedIn URL

Single-person lookups take about 10 to 60 seconds when Alpha can pin the person down.

| Ask | What happened |
| - | - |
| Full name plus company | Found in 11 s |
| First name plus company | Not found three times: the profile still showed an old employer |
| A LinkedIn URL | Found at once, and Alpha explained the employer mismatch |
| A name with nothing else | Alpha asked for the company, role or LinkedIn URL |
| A misspelt surname plus company | Alpha found the likely person under the right spelling and said so |

Profiles lag real life, sometimes by weeks or months. A recent move may not show yet.

```text Write it like this wrap theme={null}
Tell me about https://linkedin.com/in/…
```

Or:

```text Write it like this wrap theme={null}
[full name] from [company]
```

## Paste example profiles in the same message

If you have two or three people who look right, paste their LinkedIn links in the message that asks for the search, up to 10. Alpha reads them first, and it won't save a list that leaves any of them out.

**Real example: founding team members from unicorns.**

> "I'm looking for profiles of founding team members (first 10-15 people) from unicorn companies in India, ideally individuals who have left those companies and have not yet started their startup. Here are a couple of example profiles."

This gave 137 people, with those in transition, in stealth or angel investing ranked first.

Examples pasted in an earlier message don't count toward that check. Put them with the ask.

## For a team map, ask for everyone

Alpha is built to find people to hire, so by default it polishes a list toward the person you described. For a team map you want everyone, not a shortlist, and you want it grouped.

```text Write it like this wrap theme={null}
Everyone currently at [firm] in investing roles, grouped by seniority. Don't drop anyone who currently holds an investing title.
```

Say "current" or "current and former" every time.

## Ask for a number when you want a shortlist

Short follow-ups that name a size work well:

* "Can we narrow it down to top 10-15 folks" → "Curated the top 15 …" in 35 s.
* "Give some weightage to folks working across top tier LLMs in India, have some proactive presence across socials, & have north of 6-7 years workex" → the top 15 re-ranked in 58 s.

A size buried inside a long, open brief ("think creatively …") works far less well. Very open briefs take longest of all. Break them into a clear first ask plus follow-ups.

## How long does each kind of ask take?

| Kind of ask | Typical time |
| - | - |
| Product question, call prep, connecting an app | Under 30 s |
| One person, a CSV (A spreadsheet file that opens in Excel or Google Sheets.) export, a refinement | 15 s – 1 min |
| A clear search, such as company plus dates or the team at one firm | 1–3 min |
| A job description or a fuzzy brief | 3–10 min |

You can leave the page while Alpha works. **Stop** keeps the partial answer. Long, open runs also use the most of your budget; see [what uses my budget?](https://guide.kello.ai/alpha/usage-limits-and-trust.md#what-uses-my-budget)

## How do I check a result?

Copy any of these after a list comes back.

```text Count looks low wrap theme={null}
Why is this number so low? What did you leave out?
```

```text Count looks high wrap theme={null}
Why is this number so high? What did you let in?
```

```text Who is on the list wrap theme={null}
Why are these people on the list?
```

```text Check an assumption wrap theme={null}
What did you assume about [word from my ask]?
```

```text Who was rejected wrap theme={null}
Who did you consider and reject, and why?
```

For more copy-ready asks, see the [prompt library](https://guide.kello.ai/alpha/prompt-library.md).

## Related

> [card] [Finding people](https://guide.kello.ai/alpha/finding-people.md)
>
> Describing people, JDs, examples and refining.

> [card] [Prompt library](https://guide.kello.ai/alpha/prompt-library.md)
>
> Templates and real asks you can copy.

> [card] [Working with results](https://guide.kello.ai/alpha/working-with-results.md)
>
> Review, remove, save, share and export.

> [card] [Usage, limits and trust](https://guide.kello.ai/alpha/usage-limits-and-trust.md)
>
> What Alpha can't do, and what it offers instead.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.