agentic_ai_platform
Actual commits across 30 days
Independent scale: 0 → 272 commits/day; color intensity cannot be compared across tracks.
DIGITAL TWIN LAB / UNDER CONSTRUCTION
Give research a memory.
Give improvement evidence.
I’m building a digital twin for peptide research: a research harness for methods, experience and judgment, and a knowledge Wiki that connects evidence. Its growth should be visible in real changes, day by day.
TWO PROJECTS / REAL DEVELOPMENT FOOTPRINTS
See the work, day by day.
Two date-aligned tracks record actual commits. Development activity and scientific capability are measured separately.
2646
knowledge_wiki
Actual commits across 30 days
Independent scale: 0 → 2 commits/day; color intensity cannot be compared across tracks.
The full record is visible. Replay moves through actual dates; it does not simulate live growth.
UTC Partial day
agentic_ai_platform21commits
knowledge_wiki0commits
—: outside coverage / unavailable; 0: observed with no commits. Today is a partial day. Select any cell by pointer or keyboard.
Source observed at: · UTC
Non-merge commits are counted from the full history of the server default-main snapshot, using committer time. PR data is not connected; missing values are not counted as zero.
CAPABILITIES / PUBLIC EVIDENCE
Progress, with evidence to inspect.
Up to three recent public records: what changed, its validation and the next step. Development activity and capability gains are kept distinct.
Record scope
2026-10-08 · UTC · 2026-10-09T20:15:32.723537+00:00
RESEARCH / FOUR DIRECTIONS
Existing methods remain open to testing.
Hit Identification
I study how activity signals can point to candidates worth testing, and why similar peptide sequences can behave differently.
Related workAMPCliff ↗
Lead Optimization
I investigate how small changes in sequence and structure affect binding, so each round of lead optimization has a basis for comparison.
Related workPepDDG ↗
Multi-objective Optimization
I explore trade-offs among activity, selectivity, stability and developability, seeking compromises that can be explained.
Foundation Models
I study transferable representations learned from molecules and peptides, supporting prediction, screening and design while testing the limits of transfer.
Four connected research directions. Foundation models provide representation support; these are not fixed sequential stages. Computational results still need experimental testing.
Read the research methods & results ↗KNOWLEDGE / PUBLIC RESEARCH MAP
Connect questions, methods and evidence.
This brain-shaped map curates conceptual connections from public papers. It is an entry into the research, not a private Wiki topology or a measure of knowledge growth.
PUBLIC PAPERS · CONCEPT CONNECTIONS
The brain layout and connections are curated from public sources. They show conceptual relationships, not private knowledge topology or growth.
PUBLIC KNOWLEDGE / SELECT A NODE
Peptide intelligence
How can representations connect peptide chemistry, structure and function? This is the organizing question of this public map.
PepLand · 2023 author manuscript ↗Public sources checked: 2026-10-08 UTC
Curves connect representations with chemistry and structure, and activity or structure evidence with provenance. A concept connection does not imply implemented algorithm integration.
All nodes & traceable sources
- Peptide intelligence ↗
How can representations connect peptide chemistry, structure and function? This is the organizing question of this public map.
- PepLand ↗
Atom and fragment views accommodate canonical and noncanonical peptide chemistry.
- PepHarmony ↗
Contrastive sequence–structure alignment during training; sequence-only downstream inference.
- AMPCliff ↗
Small sequence changes can have large activity differences. Dedicated benchmark splits expose a difficult failure regime.
- Macrocycle evidence ↗
Public RFpeptides crystal structures provide structural controls for peptide-design analysis, including GAB_D23–GABARAP (9HGD).
- PepDDG ↗
Information-channel decomposition for peptide–protein binding ΔΔG prediction. Author-authorized V4 method and results are shown on the research page; accepted status is author-confirmed.
- Evidence → knowledge ↗
An editorial connection: preserve sources, uncertainty and tested limits so future reasoning can build on them. A design goal, not a measured learning curve.
INTELLIGENCE FLYWHEEL / TWO LOOPS
Improve research, and how research improves.
Experimental feedback takes time and resources. I want experience to become a starting point for the next investigation sooner: one loop improves peptide design, the other improves the system producing it.
Question → design → controls & evidence → revision
Outcomes and failures inform the next judgment.
Experience → gaps → method changes → independent evaluation
Evaluated methods return to the research loop.
This is the direction of the work. Releases still require human review; an autonomous research and merge loop has not been established by runtime evidence.
Explore how the cognitive system retains experience
Experience
Retain evidence, outcomes and failure boundaries.
Knowledge
Turn experience into sourced, retrievable knowledge.
Diagnosis
Identify method gaps and evidence-backed improvements.
Optimization
Propose versioned changes to skills, tools and workflows.
Evaluation
Independently check changes under fixed conditions; evaluate the evaluator too.