Utalkia

Case studies

The proof, with numbers.

Each engagement listed here is real, named or sector-anonymized at the client's request. Filter by what matters to you. Reference calls available with lighthouse clients on request — we don't list them publicly without consent.

3 of 3 engagements
Hedge funds

Hedge Funds / Asset Managers

Top 1% globally (OpenAI rating)
  • LLM orchestration
  • Agentic systems
  • Audit-grade logging
Problem

The fund's analysts were drowning in unstructured research — earnings calls, regulatory filings, broker notes — and the portfolio team wanted a reasoning system that could draft, defend, and revise investment theses without the analysts losing control of the process.

Approach

We built a reasoning model that decomposes a thesis into testable claims, retrieves supporting and contradicting evidence from the fund's own corpus, and surfaces the disagreements rather than papering over them. Every step is traceable. The model proposes; the analyst disposes.

Stack
Custom multi-step plannerHybrid RAG over filings + transcriptsAnthropic Claude (primary)OpenAI o-series (verifier)Vector DBAudit log pipeline
Outcomes
OpenAI global ranking
Top 1%
Thesis prep time
−80%
vs. analyst baseline
Coverage
2.3× more names

“It's not a wrapper. It's the way our team thinks now — only faster, and with the receipts attached.”

— Head of Research
Due diligence

A top-25 US background-investigation firm

4 hours → under 1 minute, end-to-end
  • Document intelligence
  • RAG
  • Custom workflow engines
  • Audit-grade logging
Problem

Investigators spent four hours per file: manual review, manual extraction, manual cross-referencing across thirty-plus document types. Tech firms ten times our size had told the client what they wanted wasn't possible at their price point.

Approach

We treated the work as pattern recognition over structured-and-unstructured documents and built a pipeline that extracts, classifies, and cross-references end-to-end — with full audit trails for every assertion. Investigators now review exceptions, not files.

Stack
OCR + layout-aware parsingDomain-specific extraction modelsEntity-resolution graphWorkflow orchestratorImmutable audit log
Outcomes
Per-file review
4 hrs → <1 min
Throughput
+12×
Auditable trail
100% of assertions

“Three software shops told us no. Utalkia had a working POC in six weeks and a production system in six months.”

— COO
NFT / Web3

Galileo (creator platform)

On-chain logic across 4 EVM networks, non-custodial
  • Smart contracts
  • Custom workflow engines
Problem

Creators wanted to mint and manage NFTs across multiple chains without learning the chain-specific tooling, and the team wanted on-chain logic that could enforce royalties and provenance without a centralized backend.

Approach

Aghasi built the platform from scratch — multi-chain smart contracts, a creator-friendly UI that abstracts wallet plumbing, and an on-chain provenance graph. No middleware silos. No off-chain custody.

Stack
Solidity + VyperERC-721 / ERC-1155 + custom royalty extensionEIP-712 typed signaturesMulti-chain deployment (EVM family)TypeScript SDK
Outcomes
Chains supported at launch
4
Mint time
−50%
Custody risk
Zero (non-custodial)