Backed by Y Combinator

Custom agentic AI for the world’s hardest problems.

We design, build and run AI agents for billion-dollar companies: support QA that grades every conversation, knowledge bases that catch their own contradictions, and compliance checks that cite the exact policy line.

Book a call See the work
agent/support-qa agent/kb-sync agent/policy-check
04 · Shopping assistant
Florist storefront with the shopping assistant recommending three bouquets
02 · Knowledge base
Knowledge base review dashboard with five relationship types
05 · Field inspection
Inspection report comparing SOP reference photos with the cleaner’s photos
03 · Compliance
Compliance checks citing exact numbered policy lines
01 · Support QA
support-qa / conversation review
Support QA demo: the agent refunds $408 against a $349 approval, and the scorecard flags it 6% compliant
Flagged article pair with the conflicting lines
Automatic decision approved under a company rule
WhatsApp chat where the agent answers a cleaner, citing the SOP
Florist shopping assistant on a phone recommending a bouquet
Fig. 01 — Five agents we’ve built: support QA, knowledge base, compliance, shopping assistant, field inspection

AI agents & POCs
for teams at

DoorDash Gusto Ascott Xendit tiket.com The Great Room OGCS Global Sourcy petite fleurSingapore

Y Combinator-backed

Power to the Brand is backed by Y Combinator, the accelerator behind Airbnb, Stripe, DoorDash and Coinbase.

Enterprise-proven

We have built agentic AI systems and POCs for billion-dollar companies including DoorDash, Gusto, Xendit, Tiket.com and Ascott Hotel Group.

Working with top brands

Through Sourcy.ai we work with Shein, TikTok Shop and Meiyume, whose clients include L’Oréal, Estée Lauder, Chanel and Dior.

[01]What we build

Agents that do the work, with the evidence attached.

01

Customer support QA

Every support conversation, graded.

Agents grade every conversation against your own QA scorecard and policy handbook, so your QA team reviews the flags instead of a random sample.

  • Your own criteria: customer intent, emotions, ease of interaction, menu confusion, safety issues.
  • Every finding cites the policy rule behind it.
  • From 3–5% of conversations reviewed to 100%, about 100,000 a month.
  • Graded in under half a second, for a fraction of a cent.

In this run, an agent refunded $408 against a $349 approval. The check scored it 6% compliant and quoted the refund-authority rule, in 354 ms, for under a fiftieth of a cent.

Graded · 354 ms
Scorecard header: graded in 354 ms, LLM call cost $0.000183
Flagged · 6% compliant
Financial actions comply with policy: 6% yes. Refund has prior authorization: 5% yes
Cited · Rule R1
Policy rule R1, Refund authority, cited by the failing checks
Fig. 02 — Recorded demo run on a fictional conversation
02

Self-maintaining knowledge base

A knowledge base that catches its own contradictions.

Agents read every article in your knowledge base and find the ones that duplicate, contain or contradict each other.

  • Sorts every pair: same as, subset of, overlaps with, distinct from, or conflicts with.
  • Each flag comes with a confidence score, an explanation and the exact lines that clash.
  • Side-by-side comparison, then one verdict from your content team.
  • Catches contradictions before they mislead an agent or a chatbot.
kb-review / document relations
Knowledge base review dashboard: documents sorted into five relationship types
Conflict found
A flagged article pair with its explanation and the conflicting lines
Relationship graph · conflicting lines
kb-review / selected document
Selected document with its relationship graph, a SAME_AS match at 86%, the explanation, and the conflicting information between the two documents
Similarity · confidence per pair
Conflicts dashboard: total relationships, relationship types, average similarity, and each document pair with its type, confidence and explanation
Fig. 03 — Article content shown is illustrative
03

Agents that check contract & policy compliance

Workflow compliance checks that show their working.

Every claim, invoice or document is checked against the policies and contracts that govern it.

  • Each finding links to the exact policy version and line numbers.
  • Routine cases close automatically, but only under rules you’ve approved.
  • Everything else reaches a reviewer with the evidence already gathered.
  • Every decision exports as an audit record.
Exact policy lines
claims / policy source
Each check the agent ran, with the exact numbered policy lines it relied on
Auto-decided · approved rule
Automatic decision: approved under an approved company rule
Fig. 04 — Expense claims checked against company policy
Run trace · from claim to decision
Run trace: a supplier invoice moves from reading the claim, through nine cited policy checks and the approved-rule check, to an automatic decision
Reviewer queue · conflicting evidence
Reviewer queue: an invoice that exceeds its purchase order, with the conflicting evidence, the policy line and the decision controls
Decision record · audit export
Decision record: every outcome with who decided it and the cited policy lines, one row expanded into its audit record

Fig. 04b — Illustrative screens with sample data, drawn from how the product works

04

Shopify shopping assistant

A shop assistant that checks every answer.

Shoppers ask questions and add to cart straight from the chat. Ask for a birthday bouquet under $80 and it narrows 445 arrangements to three, with prices and product cards.

  • Every reply goes through multi-layer checks against each product’s own information.
  • Unsupported claims are removed and automatically rewritten until accurate.
  • Guardrails against bot hacking and prompt injection.
  • SOP and FAQ changes take effect in real time.
storefront / ask fleur
Florist storefront with the shopping assistant recommending three bouquets with prices
Source conflict
Source conflict: the homepage and FAQ give different same-day delivery cutoffs
Behind the chat · Support inbox
support inbox / conversations
Support inbox: Fleur recommends three anniversary bouquets with snapshot prices, and the team can take over the conversation
Handled by Fleur → take over anytime
Behind the chat · FAQ & SOP editor
fleur-knowledge / faqs
Fleur knowledge dashboard: an FAQ with its main question, alternate questions in English and Chinese, and the answer, with the live published version shown
Publish → live on the next chat
Fig. 05 — Storefront demo on a Singapore florist’s public catalogue, with the support inbox and FAQ & SOP editor behind it
05

Field inspection on WhatsApp

Proof of work, checked against the SOP.

Crews report each job with photos in the WhatsApp chat they already use.

  • Every photo is checked against the exact SOP version pinned to the job.
  • SOP questions are answered on WhatsApp, with the rule cited.
  • Specific retakes are requested when a photo falls short.
  • Supervisors sign off a traceable record in seconds.
SOP v2.1 reference vs photo
inspection / room 1206
Inspection report for one room: each SOP reference photo beside the cleaner’s photo, signed off by a supervisor SOP Submitted
WhatsApp
WhatsApp chat: a cleaner asks about guest luggage and the agent answers, citing SOP v2.1
Fig. 06 — Demo data; WhatsApp shown in the demo’s simulator

Also built

Lead generation and researchAscott Hotel Group

RFID digital name cards, physical cards includedThe Great Room

Reverse chatbots that negotiate with suppliersSourcy.ai

Social media trend analysis agentsSourcy.ai

[02]Clients

From startups to billion-dollar enterprises.

Sorted by valuation

Client Valuation Industry What we built
DoorDash ~US$80B1 Food delivery Customer-support QA agent that grades tickets against DoorDash’s own QA criteria and finds the policy documents relevant to each ticket. Knowledge-base deconfliction and cleanup.
Gusto US$9.3B2 HR & payroll software AI agents that self-maintain the knowledge base
Ascott Hotel Group US$9.2B3 Hospitality Automated lead generation and research
Brex US$5.15B7 Corporate cards & spend management Knowledge-base deconfliction and cleanup for the support QA team
Xendit ~US$3B4 Payments / fintech Customer-support QA agent that grades every conversation against Xendit’s QA scorecard. Knowledge-base cleanup that finds overlapping documents saying the same thing slightly differently.
Tiket.com Blibli Tiket ~US$2.5B5 Online travel Customer-support QA agent that grades every conversation against Tiket.com’s QA criteria
The Great Room by Industrious US$800M6 Flexible workspace RFID digital name card system, including sourcing and custom-printing the physical RFID name cards in China
CoSauce — Customer support outsourcing Customer-support QA agent that grades every conversation its outsourced support teams handle
OGCS Global — Oil & gas Automated legal contract checks as part of the change-order workflow
Petite Fleur Singapore — Online florist Shopify AI shopping assistant that answers questions and adds to cart from chat. Real-time SOP & FAQ changes.
Sourcy.ai Current — Sourcing & supply chain
An AI sourcing platform for brands.
US$33M+in sourcing 390+brands 30+countries
  • Reverse chatbots that chat with suppliers on brands’ behalf
  • Agentic supplier validation
  • Internal ops agents
  • Automated lead generation and research
  • Social media trend analysis agents
Sourcy works with
SHEIN TikTok Shop Meiyume
Meiyume’s clients include L’Oréal, Estée Lauder, Chanel and Dior.

Latest public figures. 1 Market cap, 14 Oct 2026 (US$79.7B). 2 June 2025 tender offer. 3 Market cap of parent CapitaLand Investment, 14 Oct 2026 (S$11.8B). 4 2022 funding round, as reported. 5 Market cap of parent Blibli (Global Digital Niaga), 14 Oct 2026 (Rp44.2T). 6 Industrious, at its CBRE acquisition, Jan 2025. 7 Capital One acquisition, closed Apr 2026.

[03]How we work

Real operations problems. Real outcomes.

We build AI for the everyday work that keeps a business running, and judge it by what it changes, not by how clever it looks.

  1. 01 / OPERATIONS

    Real-world operations problems

    Support queues, knowledge bases, claims, field inspections. We start where the work is slow, costly or error-prone today.

  2. 02 / YOUR WORKFLOWS

    Built on your workflows, people and data

    The agent fits the way your team already works, and learns from your real conversations, documents and policies.

  3. 03 / ALONGSIDE YOU

    Built alongside you, every day

    We ship early and take your team’s feedback daily, so the agent keeps getting better at your work, not someone else’s.

  4. 04 / OUTCOMES

    10x outcomes, not AI for AI’s sake

    We use AI where it multiplies results, like reviewing every conversation instead of a small sample, and leave it out where it doesn’t.

Daniel Chua at the Google Cloud Startup Summit in Singapore
Fig. 07 — At Google Cloud Startup Summit, Singapore

[04]Founder-led

Daniel Chua

Founder, Power to the Brand

Daniel builds custom agentic AI for the world’s hardest problems. He has built agentic systems for DoorDash, Gusto, Xendit, Tiket.com, Ascott Hotel Group and Sourcy.ai.

Every engagement is scoped and built by Daniel himself. The person you brief is the person writing the agent.

[05]Contact

Bring us the problem off-the-shelf AI couldn’t solve.

Tell us what’s breaking. We’ll come back with how an agent would handle it, and what it would take to build.

Book a call