About Inflection Weekly
Inflection Weekly is a weekly research bulletin on human behavioral analysis for early mental health detection. Each Friday it surveys the past seven days of peer-reviewed papers, preprints, clinical trial registrations, regulatory developments, and industry news across nine domains:
- AI / ML for psychiatric condition detection
- Wearable biosensors (HRV, EDA, accelerometry, sleep)
- Speech and vocal biomarkers
- NLP and text-based detection
- Digital phenotyping (passive smartphone sensing)
- Multimodal fusion
- Facial expression and computer-vision affect recognition
- Ethics, regulation, and clinical translation
- Industry and product news
Every issue is deduplicated against every previous issue and against the foundational Baseline edition — no recycled content, no filler. If a domain has nothing new in a given week, that section is omitted entirely.
From system behavior to human behavior
The site is written by Isuru Gunarathne, a software engineer and researcher whose background sits at an unusual intersection: distributed systems and behavioral inference.
For the past several years Isuru's work has focused on system behavioral analysis — specifically, predicting how Kubernetes pods behave under load. His research, "A Self-Adaptive Framework for Predictive CPU Resource Allocation During Container Startup in Kubernetes" (presented at the 18th IEEE/ACM International Conference on Utility and Cloud Computing), studies how a container's behavior during the first few seconds of life can be used to forecast how it will behave under sustained load — and to allocate resources accordingly, before contention bites.
That work shares a methodological skeleton with everything in this newsletter:
Read an early signal of a state change before the state change happens.
Whether the system is a workload approaching CPU saturation or a person approaching a depressive episode, the question is structurally the same. Both fields lean on the same four ideas: high-frequency behavioral signals, model-driven inference, careful attention to confounders, and the gap between in-lab benchmark accuracy and real-world deployment. Reading psychiatric AI papers after years of Kubernetes scheduling work is uncannily familiar — the same arguments about feature leakage, the same anxieties about cross-site generalisation, the same tension between latent variables and observable proxies.
Inflection Weekly is the public-facing artefact of that crossover. It is opinionated about methodology, generous with sources, and unsentimental about what does and doesn't translate from the bench to the clinic.
How issues are produced
Each Friday at 6:00 PM UTC, an automated research pipeline:
- Reads the four most recent issues to build a deduplication index of titles, DOIs, technologies, named architectures, clinical trial IDs, institutions, and key findings.
- Runs targeted searches across PubMed, medRxiv / bioRxiv, Nature Digital Medicine, JMIR, npj Mental Health Research, ClinicalTrials.gov, and selected industry sources for the past seven days only.
- Discards anything already covered. Keeps anything genuinely new (and labels meaningful updates as such).
- Drafts the issue as a Markdown post, runs quality checks, and publishes here.
The pipeline is a Claude scheduled task running a custom skill. If you spot an error or have a paper that should be on the radar, the contact link below works.