Research Project Roots

Source Summary

This source note records a read-only survey of Vipin’s active research projects under D:/Research, especially analog-agent, uncertainty, truce-rec, tgl-rec, donebench, and uncertaintyprotein-ai4s / protein-optimization-feedback-shift.

EXTRACTED: D:/Research/Agent-AI4EDA/analog-agent is the mature AI4EDA / analog-agent project root.
EXTRACTED: D:/Research/Uncertainty is the current main root for task-grounded uncertainty in LLM-based recommendation.
EXTRACTED: D:/Research/Uncertainty-LLM4Rec is an older or archival predecessor of the current Uncertainty root.
EXTRACTED: D:/Research/TRUCE-Rec and D:/Research/TGL-Rec are adjacent LLM4Rec projects using stricter same-candidate, official-baseline, and evidence-label governance.
EXTRACTED: D:/Research/DoneBench is a benchmark project for specification grounding in tool-using agents.
EXTRACTED: D:/Research/UncertaintyProtein-AI4S is the AI4S protein optimization project about uncertainty under feedback distribution shift.
EXTRACTED: D:/Research/AI for Science had no visible project files in the inspected top-level pass.

analog-agent centers on a six-layer analog design agent with specification compilation, task formalization, world-model services, planning, SPICE-backed verification, and memory/reflection.
Uncertainty centers on C-CRP, same-candidate candidate ranking/reranking, official external baselines, and task-grounded calibrated uncertainty.
TRUCE-Rec is a CURE/TRUCE uncertainty-aware generative recommendation system, currently guarded against paper claims until official-native controlled baselines and four-domain runs are complete.
TGL-Rec studies whether LLM recommenders use temporal next-need transitions rather than mainly semantic similarity/popularity.
DoneBench evaluates whether agents can infer and satisfy executable completion criteria before acting in stateful tool environments.
UncertaintyProtein-AI4S frames protein optimization as sequential uncertainty under feedback shift, where calibration and decision quality can diverge.

EXTRACTED: Project purposes and status labels are taken from README/AGENTS/docs files in the inspected roots.
INFERRED: Uncertainty-LLM4Rec is best treated as predecessor/archive relative to Uncertainty.
AMBIGUOUS: PDF-only folders under AI4EDA likely contain drafts, references, and generated reports; they need separate source-specific ingest before being treated as evidence.
UNVERIFIED: No claim was checked by running experiments or tests in the external project roots.

Several projects contain generated outputs, ignored artifacts, and server-run paths. These are status signals but not automatically paper-facing evidence.
Names containing official, paper, or final are not sufficient proof of final evidence unless the project’s own audit/provenance rules mark them complete.