Jin Kim, co-founder and Head of Forward Deployed Engineering at LinqAlpha
LinqAlpha, the New York-headquartered AI startup building intelligence tools for institutional investors, recently raised US$22 million in a Series A round anchored by AVP, Atinum Investment, and GFT Ventures, with a notably Asia-heavy syndicate including SV Investment, Mirae Asset, Samsung Securities, East Ventures, and others spanning Singapore, Hong Kong, South Korea, Japan, and India.
Founded by Jacob Choi, Subeen Pang, Jin Kim, and Hojun Choi — a team of former Goldman Sachs analysts and MIT computer science PhDs — LinqAlpha says its multi-agent platform is already used by more than 70 financial institutions across the US, Europe, and Asia, including buy-side clients such as Causeway Capital Management and Schonfeld Strategic Advisors, collectively managing over US$5 trillion in assets.
The company positions itself against both entrenched incumbents like Bloomberg and LSEG, and a crowded field of AI challengers such as AlphaSense, Hebbia, and Rogo, betting that persistent, firm-specific reasoning — not just faster search — is where the real edge lies.
Also Read: LinqAlpha raises US$22M to bring agentic AI to public-market investors
We spoke with Jin Kim, co-founder and Head of Forward Deployed Engineering about the fundraise, the company’s Asia strategy, and how LinqAlpha plans to compete.
Edited excerpts:
Your US$22M raise features an Asia-heavy syndicate — SBI, Mirae Asset, Samsung Securities, East Ventures. Deliberate strategy or following traction?
Both. Roughly half our revenue comes from Asia Pacific, so the capital base mirrors the client base. But the syndicate was deliberately built: in institutional finance, investors are also distribution. Firms like SBI, Mirae Asset, and Samsung Securities are operating institutions in the markets we serve; that alignment shortens trust-building cycles that normally take years. We didn’t raise Asian capital to enter Asia later; global coverage, Asian languages, and multi-asset support were in the design from Day One.
You’re headquartered in New York, yet clients and capital skew Asia. When does it make sense to shift your center of gravity to Singapore or Tokyo?
We, at LinqAlpha, run a distributed model rather than one centre: New York for business development, Seoul as our product/engineering hub, subsidiaries in Hong Kong and Singapore, with London next. Singapore is where we’ve made the on-the-ground commitment, a dedicated local team, not a sales outpost, but people who co-design deployments with regional institutions. The timing isn’t accidental: MAS just launched the Future of Finance Institute to move AI in financial services from experimentation to deployment. That deployment gap is our entire business.
You count 70+ financial institutions as clients, but client count can be a vanity metric. How embedded is LinqAlpha in daily workflows, and how do you measure it?
We agree it’s vanity, which is why we manage for depth across three layers:
- Daily-workflow usage: morning briefings, alerts, meeting-prep agents that fire before the user’s day starts, not ad-hoc Q&A
- Expansion within accounts: trials converting to multi-seat, multi-team deployments
- Integration depth: clients moving from app to API access, building our agents into their own systems
The pattern we watch: users going from reading our output to relying on it—starting with a briefing, pushing results to PMs, then asking us to build custom trackers for signals nobody else covers, often in Asian-language sources their other tools can’t read.
Hebbia, AlphaSense, Rogo, Dataminr — plus Bloomberg and LSEG embedding AI into terminals. Why should a CIO choose LinqAlpha over waiting for incumbents to catch up?
We’re solving different problems. Bloomberg and LSEG are indispensable data infrastructure; AlphaSense built a strong content library with AI on top. But a platform selling the same content to everyone will, by design, give every subscriber the same AI answer from the same corpus.
What a CIO competes on is the firm’s own frameworks, thesis history, and internal research. We encode that — per firm, permissioned, isolated — in what we call a second brain. Ask two funds “what are the top AI trades today” and a generic tool names the same mega-caps for both. Our platform answers in the context of each firm’s own mandate. We sit on top of data clients already license and reason across asset classes — equities, macro, credit, FX –in one connected system rather than siloed products.
AI hallucinations can directly influence capital allocation. What safeguards ensure accuracy and auditability, and has a failure ever cost a client?
We designed the platform assuming models are fallible, so safeguards are architectural. Every claim is grounded in licensed, vetted data with a citation back to source, auditable in one click. Numbers are computed deterministically through code, not generated by an LLM. We run multiple frontier models and neutralise their individual biases; our research on systematic sector/style biases in base models was accepted at ICLR and presented at a BlackRock quant conference.
Also Read: In the age of AI, people matter more than ever
We call this discipline harness engineering: as models get more powerful, value shifts to controlling them — permissioning, audit trails, human-in-the-loop checkpoints built for regulated finance. On failures: we have not had an incident where a platform error drove a capital loss for a client.
Clients feed you proprietary research and conviction signals. How do you handle data security?
Isolation isn’t a feature; it’s the product’s precondition. Each firm’s knowledge layer is siloed: notes, theses, and research never train shared models and never cross accounts. One client’s second brain is architecturally invisible to every other client’s.
For institutions wanting an even tighter boundary, our API and MCP deployment patterns let agents operate inside the client’s own environment, so sensitive context need not leave their perimeter. The incentive structure matters too: our business model is per-firm value, not data aggregation. Security reviews and regulatory addenda with global banks are table stakes, and we treat them as part of the product.
Southeast Asia is fragmented — multilingual, multi-regulatory, inconsistent data. How does the platform handle Bahasa Indonesia filings, Thai regulatory announcements, and Vietnamese commodity flows simultaneously?
That fragmentation is the inefficiency we were founded to arbitrage. The platform analyzes 20,000+ companies across 80+ markets in 20 languages, reading local-language primary sources natively rather than waiting for English translations that arrive late or never.
In Asia, that’s where alpha lives: information that’s public but not yet priced because it sits behind a language barrier. Clients already track signals in Chinese-, Korean-, and Japanese-language sources that English-first platforms structurally miss; the same architecture extends across Southeast Asia. Equally important is multi-asset design: an Indonesian commodity signal reads through to Singapore-listed equities, regional FX, and credit in one connected graph. Where coverage needs deepening, we build it hand-in-hand with regional clients—that’s the global best practice we’re bringing to Singapore, not a US product with a Singapore price list.
With a Berkeley MFE/Goldman/MIT pedigree, doors open easily, but institutional adoption is slow. What’s been the biggest obstacle converting pilots to long-term contracts?
Never model quality in a demo. The real obstacle is earning a place in daily workflow at a conservative institution—what kills pilots industry-wide is a tool that impresses in week one and is forgotten by week six. We treat every trial as an implementation, instrumenting adoption user by user and workflow by workflow.
The second obstacle is institutional trust: security review, compliance sign-off, data governance. We stopped treating that as friction and started treating it as the sale, because the risk owner is usually the real buyer. What converts pilots is co-designing an AI roadmap with client leadership over the next one to two years, rather than selling seats.
Buy-side clients manage US$5 trillion+ in assets, striking for a Series A. What’s your pricing model? Recurring SaaS, or a services-heavy business that doesn’t scale?
It’s recurring software by design: seat- and entitlement-based subscription with enterprise tiers for API access and advanced modules. The same platform serves a hedge fund pod and a bank’s research floor. The insight most people miss: the “bespoke” part—learning each firm’s framework—is performed by the system itself. The second brain is built by agents from the client’s own permissioned data, not consultants billing hours. That makes personalization compound instead of costing more. The AUM figure is a statement about who trusts us, not a revenue multiplier—but reference clients at that tier are the moat, because institutional buyers follow institutional proof.
If AI “changes what analysts can know,” doesn’t wide adoption commoditise the very edge you promise?
That critique is fatal for generic AI. This is exactly why we built the opposite. If every investor used the same model on the same data, the edge would be arbitraged away in a quarter. Our architecture inverts it: agents reason in the context of each firm’s own thesis history and mandate, so two funds asking the identical question get different, both correct, answers—the platform amplifies different brains. Adoption doesn’t converge outputs; it compounds each firm’s accumulated judgment.
Also Read: Is generative AI the game-changer for productivity?
Think of Bloomberg in the 1990s: everyone had the terminal, yet returns diverged wildly, because the edge was never the tool; it was what each firm did with it. We’ve made “what each firm does with it” the product itself. AI is shifting the scarce resource from information access to quality of questions and speed of connecting dots. The real risk for a CIO isn’t adopting AI too early; it’s letting a competitor’s second brain start compounding a year before yours does.
The post Most AI pilots die in week six. Here’s what LinqAlpha does differently appeared first on e27.