Saw S. Lin (張智奇)

A fellow human being — just curious, committed, respectful.

Building products end to end, designing AI-powered systems, and researching LLM inference — capable on my own, and a natural collaborator.

Portrait of Saw S. Lin
About
  • Finishing an MSc in CSIE at National Taiwan University — researching LLM inference acceleration.
  • Independent builder — clients span institutions and businesses, including STPI @ NIAR, a Taiwanese science-and-technology policy research institute, and IFC Travels & Tours.
  • 6+ years across product, project, and engineering. Angel-invested in SleekEV during its bridge round.
Selected Work

First-author paper (arXiv, 2026) from my research at NTU — an industry–academic collaboration with Delta Electronics — under Prof. Jyh-Shing Roger Jang. It builds a conditional, best-first draft tree for speculative decoding, no training required, hitting up to 6.6× speedup over autoregressive decoding on Qwen3-4B with the highest average acceptance length among the methods we evaluated. Open-sourced.

Read paper
ResearchAlgorithm

I designed Seek end-to-end at STPI @ NIAR — the recommendation algorithm, the API, and the front end — on the center's proprietary scholar database. At heart it's a scholar-matching engine: name a need and it surfaces the right academics. Its first job is helping evaluate government research grants: for each application, a discipline convener — the senior scholar who runs a field's grant reviews — has to line up reviewers, and as of the 2026 review season about 3.4 of their 5 picks come straight from Seek on average across 180+ applications, turning a search that took days into minutes for each application. The same engine can just as easily match scholars to conference panels, or companies hunting for the right academic partner. It's the intelligence inside TOP Grants, the workspace those conveners use to run their reviews.

AlgorithmProductAPI DesignFrontend

For these grants, a senior scholar in each field — the discipline convener — decides who reviews each application and gives the final verdict. The old system made that a slog — a 2000s-era back-office UI: two look-alike sites for the two kinds of applicant (PhD candidates and early-career scholars), no way to see where any application stood, and initial reviewers invited one email at a time, by hand. I designed and rebuilt the convener's side of it in Flutter Web — one app, one URL, both applicant types with live status at a glance. The convener now does just three things — choose the initial reviewers, read the reviews that come back, and give the final verdict — while the busywork falls away to the support staff who run the back office, and each application, its attachments, and every review sit on a single page instead of scattered across several screens. Seek even suggests those initial reviewers, so finding them drops from days to minutes.

FrontendProductAPI Design

An AI patent-and-literature analysis platform I shaped at STPI @ NIAR — product design, scope, API specification, and the front-end build. It renders a technology-function matrix that maps where a technology is crowded or wide open, giving industry and academia a fast read for innovation assessment and R&D direction.

FrontendProductAPI Design
Products

getScholarshipTW

Launching this month

A platform that helps students in Taiwan find — and actually win — the scholarships they're eligible for.

Experience
  1. 2025.07–Now

    At STPI — a science-and-technology policy research institute in Taiwan — I design and build its digital products end to end. Three have shipped: Seek (a scholar-recommendation engine) and TOP Grants (the review workspace for discipline conveners — the senior scholars who run each field's grant decisions) — which together modernize how the government reviews its research grants — plus TEMatrix (patent-and-literature analysis), which maps where a technology is crowded or wide open for industry and academia. I owned Seek's algorithm and led product, API, and front-end design across all three, working shoulder-to-shoulder with the back-end engineer and directly with the client — who's become a strong advocate. Rebuilding the conveners' workspace was the highest-stakes piece: those conveners are some of Taiwan's most respected scholars, and a clumsy tool reflects straight back on the office running the program. What I'm most excited about is where Seek goes next — growing it from a reviewer-matcher into the connective tissue of the research ecosystem: pairing students with advisors, and academics with the companies that need them.

  2. 2025.01–06

    Part-time AI product strategist for an AI virtual-character (VTuber) startup. I authored a complete US$6M fundraising pitch for the JETRO deep-tech program, and built an automated content pipeline — HeyGen, VTube Studio, generative AI, and Python — that turned work of days, sometimes weeks, into hours, producing 6+ video assets. I also ran competitive intelligence and product-strategy consulting across the health-supplement and semiconductor verticals.

  3. 2024.02–07

    Spacematter Apps is a studio started by a close friend, where I collaborate as a founding member. I co-designed the ShiShi Timer app with one of the founders, and on the IFC Travels & Tours engagement (Yangon) I was the client-facing lead — pitching, presenting three strategy proposals, and running the relationship end to end — while the Spacematter team built the design and prototype: an MVP integrated with one of Myanmar's largest mobile-payment super-apps, with feasibility vetted alongside Huawei's China team.

  4. 2023.08–2024.02

    As IT project manager I ran six concurrent projects — remotely from Myanmar for a Singapore app studio — for clients including Isetan Singapore, AIA Myanmar, and Extra Space Asia.

  5. 2022.10–2023.02

    A self-funded venture and my first real turn as a founder: I set out to validate a food-education app to help people with diabetes live better, starting from nutrition data for 300+ local foods — testing the value before building the product.

  6. 2018.07–2022.05

    Over four years I grew from project manager to technical product manager, owning the roadmap for enterprise people-operations platforms — the systems that run payroll, performance, and leave for whole companies. I grew the client base to six enterprises handling 2,700+ employees a month, and the platforms I shipped delivered real gains for those clients: 133+ staff-hours saved a month, performance appraisal completion up to 94%, and leave-approval waits cut by ~80%.

    Two earlier projects stand out. On a nationwide government tender for a digital-cattle system, we were the local underdog against far better-funded players — foreign firms included — and still finished one of two finalists among 16 teams (our prototype tested 144 cattle in three days). And on a microfinance platform, I led the team to harden system stability, tune it to real customer needs, and bring reporting up to industry best practice — work that helped recover about US$100k in bad debt.

  7. 2018.03–05
    SKYBIT.ASIASoftware Development InternSingapore · Remote from Myanmar
Say hello

Whether you're building something, backing something, or simply want to make friends — say hello.