Engineering Behavior, Building Systems.
Every system starts as a position, and every position is a claim about behavior.
Psychological scientist · Systems builder · Consultant
I study why people do what they do, then build the systems that account for it. Experimental psychology on one side, shipped software on the other — the same discipline pointed at two ends of the same problem.
A full consulting engagement for the Musicity platform: from raw user interviews through financial structuring to a business model built to force product-market fit rather than wait for it.
End-to-end strategy for a music platform chasing aggressive fit
Structured interview protocol run across artist and listener segments. Coded transcripts turned unstructured complaints into a ranked list of jobs-to-be-done, separating what users said they wanted from what their behavior demanded.
Automation pipelines, native apps, and assessment platforms — every one of them starts from a behavioral claim and ends in an interface that has to survive contact with a real user.
Custom n8n pipelines that capture CRM signal from calls and emails and land it in Pipedrive automatically — with manual-review safeguards on every branch where a wrong write would be worse than no write.
Custom n8n pipelines that capture CRM signal from calls and emails and land it in Pipedrive automatically — with manual-review safeguards on every branch where a wrong write would be worse than no write.
| Step | Role | Sends to |
|---|---|---|
| Call recording | capture source | n8n trigger |
| Email thread | capture source | n8n trigger |
| n8n trigger | orchestration trigger | Entity extraction |
| Entity extraction | transform step | Confidence gate |
| Confidence gate | routing decision | Pipedrive (high confidence); Manual review (ambiguous) |
| Manual review | human-in-the-loop hold | Pipedrive (approved write) |
| Pipedrive | system of record | terminal |
A burn-rate ticker for attention. Costly anchors distraction time directly to a personal savings goal and runs the number in real time — because the stakes in a user's environment are never actually zero.
€0.0000
Priced at €18.50/hour against the goal. The meter has no zero state — that is the entire argument.
The behavioral truth the app is built on: financial stakes in user environments are never completely zero. Costly makes the number visible instead of letting it stay abstract.
A relationship assessment platform, built end to end alone: instrument design, database, LLM profiling layer, and interface. Responses are scored into a psychometric profile and read back as a live radar.
Simulated readout. The query lines and the profile copy are composed on the client from the scores shown — a demo of the OopsCupid output surface, not a live model call.
| Profile | Attachment security | Conflict repair | Emotional granularity | Autonomy support | Shared meaning | Trust calibration |
|---|---|---|---|---|---|---|
| Profile P-0417 | 0.72 | 0.55 | 0.84 | 0.61 | 0.78 | 0.66 |
| Profile P-1183 | 0.41 | 0.79 | 0.52 | 0.88 | 0.47 | 0.73 |
| Profile P-2065 | 0.86 | 0.63 | 0.70 | 0.44 | 0.90 | 0.58 |
| Profile P-3390 | 0.58 | 0.86 | 0.63 | 0.72 | 0.55 | 0.81 |
A learning tracker built on a two-source data model: the canonical engineering curriculum on one axis, the coverage actually observed at universities on the other. The gap between them is the product.
The reference matrix: what an engineering discipline is agreed to contain, topic by topic, independent of any single institution.
What universities actually teach, gathered per program and mapped onto the same topic grid so the two sources become directly comparable.
24 of 60 cells sit at least 0.25 below the canonical weight of their topic. Coverage values are generated from a seeded model of the two-source matrix — the same numbers every load.
| Topic | Canonical weight | Milan A | Milan B | Turin | Bologna | Padua | Pisa |
|---|---|---|---|---|---|---|---|
| Discrete math | 0.77 | 0.66 (gap −0.11) | 0.89 (gap +0.12) | 0.69 (gap −0.08) | 0.24 (gap −0.53) | 0.40 (gap −0.37) | 0.56 (gap −0.21) |
| Algorithms | 0.89 | 0.79 (gap −0.10) | 0.93 (gap +0.04) | 0.26 (gap −0.63) | 0.62 (gap −0.27) | 0.49 (gap −0.40) | 0.87 (gap −0.02) |
| Computer architecture | 0.79 | 0.46 (gap −0.33) | 0.42 (gap −0.37) | 0.57 (gap −0.21) | 0.53 (gap −0.26) | 0.38 (gap −0.41) | 0.82 (gap +0.03) |
| Operating systems | 0.69 | 0.48 (gap −0.21) | 0.86 (gap +0.17) | 0.37 (gap −0.32) | 0.74 (gap +0.05) | 0.61 (gap −0.08) | 0.19 (gap −0.50) |
| Networks | 0.47 | 0.39 (gap −0.08) | 0.71 (gap +0.24) | 0.04 (gap −0.43) | 0.60 (gap +0.13) | 0.39 (gap −0.08) | 0.42 (gap −0.05) |
| Databases | 0.53 | 0.23 (gap −0.30) | 0.18 (gap −0.35) | 0.44 (gap −0.10) | 0.57 (gap +0.04) | 0.23 (gap −0.30) | 0.56 (gap +0.03) |
| Statistics | 0.96 | 0.86 (gap −0.10) | 0.97 (gap +0.01) | 0.86 (gap −0.10) | 0.92 (gap −0.03) | 0.97 (gap +0.01) | 0.97 (gap +0.01) |
| Machine learning | 0.89 | 0.56 (gap −0.33) | 0.79 (gap −0.10) | 0.84 (gap −0.04) | 0.34 (gap −0.55) | 0.56 (gap −0.33) | 0.21 (gap −0.67) |
| Distributed systems | 0.97 | 0.19 (gap −0.79) | 0.88 (gap −0.09) | 0.68 (gap −0.29) | 0.97 (gap −0.00) | 0.83 (gap −0.14) | 0.40 (gap −0.57) |
| Compilers | 0.78 | 0.63 (gap −0.15) | 0.73 (gap −0.05) | 0.32 (gap −0.46) | 0.61 (gap −0.17) | 0.48 (gap −0.30) | 0.86 (gap +0.08) |
A channel, a thirty-issue magazine, a thesis with a significant result, and two books. Different surfaces, one habit: make the observation, then make it public.
The character is the format. Straight-faced delivery on psychology, systems, and the parts of engineering nobody puts in the syllabus — played completely seriously, which is exactly why it works.
subscribe, I'm serious!
Not a joke. Not a bit. The call to action.
A thirty-issue digital art magazine delivered by email. Each issue ends in one-click reply buttons — the lowest-friction response surface an inbox allows — built to convert passive readers into a measurable engagement signal.
The Daily Sublime
from: sina · issue 30 of 30
One tap to reply
replies logged this issue: 000
The manuscript from the hero sequence is not a metaphor. One non-fiction book on the discipline of observation, one novel that refuses to resolve.
A working manual for looking at things properly: attention as a trainable instrument, observation as the discipline underneath every method I use elsewhere on this page.
Introspective literary fiction, drafted in the same notebook as the complexity-theory notes. A narrator trying to verify a proof about himself that he cannot construct.
An automated interview simulation testing how self-referential visual feedback changes candidate behavior. Repeated-measures ANOVA over N = 96 shows significantly greater expressiveness when candidates see themselves.
Greater candidate expressiveness under self-referential visual feedback, p = .009.
An automated interview simulation testing how self-referential visual feedback changes candidate behavior. Repeated-measures ANOVA over N = 96 shows significantly greater expressiveness when candidates see themselves.
Greater candidate expressiveness under self-referential visual feedback, p = .009.
The individual points and trajectories shown are simulated from each condition’s reported mean and standard deviation with a fixed seed; they are illustrative of the distribution, not the raw response data.
| Condition | Mean | SD | 95% CI | n |
|---|---|---|---|---|
| No visual feedback | 3.41 | 0.82 | [3.24, 3.58] | 96 |
| Interviewer-image feedback | 3.62 | 0.79 | [3.46, 3.78] | 96 |
| Self-referential feedback | 4.18 | 0.74 | [4.03, 4.33] | 96 |
MSc in Applied Experimental Psychological Sciences, University of Milano-Bicocca, 91/110