Hi, I'm Sanka's site assistant. I only know about his work, research and projects. What brings you here?
Full-stack developer and bioinformatician, building end-to-end bioinformatics platforms and user-friendly solutions.
Tallinn University of Technology · Estonia
Thesis: “User-friendly Analysis of Droplet Experiments”, supervised by Prof. Ott Scheler and Prof. Olli-Pekka Smolander, and defended in October 2023. The work attacks a practical bottleneck in droplet microfluidics. Every lab writes its own bespoke image-analysis script, and the results are impossible to compare. It delivers EasyFlow and an analytical platform that turn high-throughput droplet imaging into something a biologist can run without writing code, alongside machine learning methods for biological data ranging from supervised models to deep learning for object classification.
Uppsala University · Sweden
Thesis: “Characterisation of Diatom Silicic Acid Transporter Protein using Computational Methods”, supervised by Prof. Lucie Delemotte (KTH Royal Institute of Technology) and Prof. Parvez Alam (University of Edinburgh). It carries the undergraduate SIT1 work up a level, moving from gene sequence to the structure and behaviour of the transporter protein itself, with molecular modelling and biomaterials perspectives from the two groups. The degree paired that computational biology with the applied side of the field: biological macromolecules, bioprocessing, synthetic biology, innovation management and project work.
Universitas Gadjah Mada · Indonesia
Thesis: “Analisis In Silico Variasi Gen SIT1 pada Diatom (Bacillariophyceae)”, an in silico study of SIT1 gene variation across diatoms, supervised by Prof. Niken Satuti Nur Handayani and Prof. Eko Agus Suyono. Starting from GenBank sequence mining, it designed 27 primer pairs in FastPCR and used in silico PCR on 14 usable sequences; the selected primer and its degenerate variant resolved nine diatoms into two clusters matching frustule morphology, centric against pennate. It was the first piece of purely computational biology, and the start of the diatom thread that ran through the master's.
Icosagen Cell Factory
Work that runs the full span from web engineering projects through to machine learning and AI ones. Internal AI infrastructure serving 200+ users and cutting costs by up to €100K a year. Owns the CI/CD workflow for a growing stack of Flask and FastAPI microservices behind Streamlit and Next.js interfaces. Also builds and runs an internal hub of web apps on PostgreSQL with self-hosted S3 storage (MinIO), maintains the internal sequence analysis and antibody discovery tools, and drives workflow development for project managers. Also took the company's AI adoption problem to the University of Tartu Digital Innovation Industry Project, where a student team diagnosed why the internal AI platform was under-used and built template automation for SOPs and presentations; several of their ideas are already deployed.
Tallinn University of Technology
Machine learning and bioinformatics research alongside doctoral studies, ranging from simple supervised and unsupervised methods through to deep learning for object classification. Published 10+ articles in Q1 journals, delivered the image analysis course (IEE1820) and supervised students on the software DevOps course (ITI0303). Built EasyFlow, cutting an analysis workload from days to minutes, and maintained the virtual machine infrastructure the group's web applications ran on.
German startup · Research collaboration
Automated data quality control for genomic analysis and cloud-based bioinformatics. Built ETL pipelines for gene-bank queries and a diagnostics algorithm reaching >98% accuracy on 1M+ DNA arrays. It was also the first time I worked with Docker, and the one-liner commands turned out to be more fun than expected.
iGEM Foundation · USA · part-time
Co-founded a national synthetic biology competition that drew over 1000 competitors, then took it into the iGEM Foundation, a 60,000-member organisation, as the iGEM Indonesian League, working with the VP of League after the acquisition. Market analysis across 30+ universities shaped the growth plan that took participation from 60 to 150 in a single year.
Åbo Akademi University · Turku, Finland
Six months of bioinspiration research with Prof. Parvez Alam, studying how organisms solve mechanical problems and what materials science can borrow from them. The work spanned the pistol shrimp's dactyl plunger as a damage-tolerant sandwich composite, the surface architecture that lets the decorator crab camouflage itself, diatom frustules and the extensibility of single mucilage molecules, and molecular simulation of carbohydrates. Three of these threads became peer-reviewed papers, and the collaboration with Prof. Alam continued into the master's thesis.
JALA Tech · Indonesia
Led business development for aquaculture technology covering farm water sensors, driving market expansion and regional partnerships across Indonesia. Also represented the company at both national and international forums.
hanungchr.vercel.app · Client site
A portfolio site built for illustrator and visual storyteller Hanung Charendra: “Folk tales, tiny creatures, and everyday rituals, drawn one story at a time.” Image-led and scroll-driven, it carries three bodies of work, the Cakala picture books on the forest creatures of Mount Muria, the CobyVerse specimen universe, and the Daily Consumption sticker sheets, then hands visitors off to Instagram, Behance and a downloadable portfolio.
Platform engineering · Workflow automation
Bioinformatics tools built from internal problems, not theoretical ones. Automating analysis workflows and building platforms around them so researchers move from raw data to decisions faster. The scope is wide, from simple analysis pipelines up to high-throughput workflow development.
Python · Streamlit · Image analysis
Microfluidics, bioinformatics and image analysis tooling that cut a multi-day droplet analysis workload down to minutes. Built during the PhD at TalTech and hosted there, it is the tool the doctoral thesis is named for.
Google ADK · Multi-agent system
An intelligent multi-agent system that automates systematic literature reviews and research data extraction using Google's Agent Development Kit.
Next.js · Vertex AI
Personal gym workout logger with AI coaching, analytics, streaks and shareable metrics.
Next.js · Developer tools
A simple web tool that turns code snippets into clean, shareable visuals for docs, slides and social posts.
moatcheck.xyz · AI evaluation
A free tool that answers one question for founders: will AI commoditise your startup? Scan a URL or a description and it scores the business across several moat dimensions, walks through the reasoning signal by signal, and gives a two-year verdict on how defensible the position looks.
Streamlit · Deepgram · pydub
A modular podcast and audio cleaning tool: speech AI strips filler words, silence detection compresses the dead air, and it outputs cleaned audio alongside a transcript.
LSTM · AutoARIMA · Analytics
A prototype dashboard for biotech stock analysis, applying LSTM and AutoARIMA models to forecast movements with interactive analytics on top.
Python · Agent library
An agentic library that simplifies multi-agent workflows for research and prototyping.
h-index 10 · i10-index 10
Peer-reviewed work in microfluidics, droplet image analysis and computational biology. 250 citations across 12 PubMed-indexed articles, h-index 10.
Streamlit · FastAPI · Flask
Turning analysis scripts into tools people actually use: Streamlit apps for lab workflows, FastAPI services behind them, and legacy Flask apps for DNA sequence analysis kept alive. 15+ APIs and Streamlit apps in production.
Next.js · Tailwind · TypeScript
Oversee more than 20 web applications, APIs and microservices. The work ranges from internal database management systems for biological data to AI-powered tools for translation and proposal writing, plus a hub of bioinformatics extensions from simple CRUD apps to complex analysis pipelines. Heavier workflows run on Celery with Redis as the broker, monitored through Flower, with Portainer and a custom port dashboard tracking endpoints across the services. Front ends are Next.js and Tailwind.
Docker · Ansible · CI/CD
Containerised deployments and CI/CD pipelines on GitLab with Ansible and Docker across 15+ microservices, self-hosted S3 storage with MinIO, PostgreSQL behind a 30+ app hub, virtual machine infrastructure for research web apps, and cloud work across AWS (EC2, S3) and GCP (Compute Engine, Storage).
Vertex AI · Ollama · Open WebUI
Internal AI infrastructure saving up to €100K a year: self-hosted models via Ollama behind Open WebUI, managed inference on Vertex AI (Gemini), and multi-agent systems built on Google's Agent Development Kit. On the personal side, OpenClaw runs a local assistant named Paijo, wired into Telegram and the chat on this site, backed by Ollama models and a growing tool set. I previously tried a Hermes-based agent setup, but OpenClaw's tighter integration and local-first control fit better. For day-to-day coding I reach for Claude Code; Codex was an earlier stop along the way.
medium.com/@im-sanka
Essays on working with AI, life and hiring in the bio/tech industry, and what actually changes when new models land. Most recent: “A Birthday Gift I Never Expected” (2026-08-03).
Off-screen hours
Guitar is the hobby that has nothing to do with a keyboard. It is the one reliable way to stop thinking about pipelines for an hour.
Long-running series preferred
A steady diet of anime and series. It is the same taste for long-form world-building that made One Piece an easy card game to fall into.
Genshin Impact · Since 2020
Day-one player since Genshin Impact launched in 2020. Still exploring Teyvat, saving primogems, pretending the resin cap does not exist, and keeping 100% exploration in every region.
Started July 2026
Newly into the One Piece Trading Card Game. Deckbuilding turns out to be the same puzzle as any optimisation problem: a constrained budget, combinatorial interactions and an opponent adapting to you.
Mindstone Tallinn AI Meetup · Tehnopol
A talk on the gap between an AI demo and something people actually run. It covers why sensitive healthcare data pushes you to internal tooling with Ollama and Open WebUI, walks through two builds (this site on GitHub Pages with Copilot and Actions, and Sweatlogger on Vercel with Google auth), and argues that AI-only coding produces poor architectural decisions when the person prompting has no infrastructure knowledge: it scales badly and it costs more. The closing point is that AI reaches past the interface into systems and agentic work.
PyCon Estonia and friends
Curiosity across different worlds of programming: conferences, meetups and the people who make them worth attending.