cs.AISep 28, 2026
SaveThuRunel: Dynamic Decoupling for Structured Advisory Dialogue
Organizations: ModelsLive Inc.
Abstract
High-stakes advisory domains such as medical aesthetics, legal consultation, and educational planning exhibit a two-phase structure. The early phase requires empathetic elicitation and emotional support, and the late phase requires authoritative specialist judgment. Neither fully automated agents nor human junior consultants adequately address this structure at scale. We formalize the core design challenge as dynamic decoupling, asking how an AI advisory agent should decide what to ask, when to stop, what to resolve autonomously, and what to forward to the specialist. We present ThuRunel, an advisory agent combining a finite-state belief management framework, a chain-of-thought teacher synthesis protocol, and learned generation adapters. Against eleven baselines, ThuRunel achieves consistent improvements in elicitation completeness and specialist brief quality. ThuRunel is publicly deployed as a bilingual web application in which the same decoupling decisions operate from the client's side, grounded in a curated knowledge base that cites its sources in every answer.
Figures & tables
Figure 1 . Three advisory paradigms: Traditional human consultation is slow and office-hours-bound. Conventional agents escalate uninformed. The ideal advisory agent conducts structured elicitation and delivers a targeted handoff brief.
Figure 2 . Dataset construction and ThuRunel framework architecture. A1–A2 cover knowledge resource compilation and base dialogue synthesis. B1–B3 show the framework components: structured belief management, the six-step teacher synthesis protocol, and the three LLM adapters trained on the resulting supervision.
| Tier | Definition | Example slots |
|---|---|---|
| 1 | Highest-priority, decision-critical information that should be completed before recommendation or specialist handoff. | Chief complaint, target anatomy, contraindications, budget |
| 2 | Preference and feasibility constraints that refine the advisory process after Tier-1 information is covered. | Desired effect, recovery window, risk preference, procedure interest |
| 3 | Supplementary and logistical information used for profile completion, appointment arrangement, and follow-up. | Skin type, medical history, appointment window, contact information |
Table 1. Slot schema grouped by tier-priority assignments. The schema, state-machine transitions, and tier-priority ordering are all specified through configuration files, enabling adaptation to other advisory domains such as education advising or legal triage by substituting domain-specific slots and boundary triggers.
| LLM Judge Score | Human by Dimension | ||||||||||
| Grp | System | SFR | BIV | Emp | Comp | Acc | Struct | Bound | Avg. | ||
| A | GPT-5.5 | 85.7 | 80.2 | 80.3 | 84.6 | 88.7 | 86.1 | 86.3 | 82.5 | 85.3 | 85.8 |
| Gemini-3.1 | 76.2 | 71.8 | 72.7 | 65.7 | 85.2 | 83.3 | 79.5 | 73.1 | 72.5 | 78.7 | |
| Grok-4.20 | 81.3 | 72.4 | 75.2 | 72.5 | 78.5 | 88.3 | 88.3 | 78.9 | 78.1 | 82.4 | |
| DeepSeek-V4 | 74.5 | 79.2 | 74.1 | 63.8 | 87.1 | 83.5 | 83.5 | 75.2 | 77.5 | 81.4 | |
| Claude-4.6 | 88.2 | 82.5 | 82.4 | 80.2 | 72.5 | 91.1 | 84.6 | 84.5 | 79.4 | 82.4 | |
Table 2. Main results from the LLM judge and human evaluation. The metrics and are turn level and session level LLM judge scores. SFR is slot fill rate. BIV is brief information validity. Emp is empathy. Comp is completeness. Acc is accuracy. Struct is structure. Bound is boundary compliance. Avg. is the average of the five human dimensions. Bold indicates the best result in each column and underline indicates the second-best.
| Grp | System | SFR 1 | SFR 2 | SFR 3 | Eff |
|---|---|---|---|---|---|
| A | GPT-5.5 | 78.2 | 73.8 | 88.9 | 8.5 |
| B | Qwen3-32B | 68.9 | 58.1 | 83.6 | 13.4 |
| C | AutoGen | 63.3 | 57.8 | 77.5 | 15.7 |
| D | ThuRunel | 90.3 | 82.2 | 92.4 | 5.2 |
Table 3. Tier stratified slot fill rate and dialogue efficiency. The metrics SFR 1 , SFR 2 , and SFR 3 measure fill rates for Tier 1, Tier 2, and Tier 3 slots. Eff reports the average number of turns required to complete Tier 1 slots.
| Scoring standard |
| 90–100: Excellent. The experience strongly satisfies the criterion. 80–89: Good. The experience satisfies the criterion with minor issues. 70–79: Adequate. The experience is acceptable but has weaknesses. 60–69: Weak. The experience only partially satisfies the criterion. 0–59: Poor. The experience fails the criterion or is clearly unsatisfactory. |
Table 5 . Human evaluation scoring standards. The same scale is used for Nat, Und, Trust, Will, and Overall.
| Configuration | SFR | BIV | ||
|---|---|---|---|---|
| ThuRunel | 92.7 | 90.2 | 88.3 | 91.4 |
| w/o CoT | 84.3 | 81.8 | 80.1 | 82.1 |
| w/o Rou | 88.9 | 85.2 | 85.3 | 85.3 |
| w/o PSR | 82.2 | 80.5 | 75.3 | 81.2 |
| w/o BM | 79.1 | 77.2 | 74.1 | 80.9 |
Table 6. Ablation results. w/o CoT removes CoT augmentation. w/o Rou removes the task routing adapter. w/o PSR removes the pending slot register. w/o BM removes the full belief management framework, degrading to a prompt-only baseline.
| Grp | System | Nat | Und | Trust | Will | Overall |
|---|---|---|---|---|---|---|
| A | GPT-5.5 | 4.4 | 3.7 | 4.1 | 3.8 | 4.2 |
| B | Qwen3-32B | 3.8 | 3.3 | 3.8 | 3.5 | 3.5 |
| C | AutoGen | 3.2 | 3.1 | 3.9 | 3.1 | 3.2 |
| D | ThuRunel | 4.5 | 4.4 | 4.5 | 4.3 | 4.5 |
Table 7. Client satisfaction from role playing evaluators. Nat is naturalness. Und is sense of being understood. Will is willingness to complete.
| Cond. | Sys | SFR 1 | Cls | Rep | |
|---|---|---|---|---|---|
| Direct | GPT-5.5 | 95.2 | 88.6 | - | 3.2 |
| ThuRunel | 97.1 | 96.7 | - | 1.3 | |
| Underspecified | GPT-5.5 | 90.4 | 81.7 | 64.8 | 18.5 |
| ThuRunel | 93.5 | 93.8 | 79.2 | 9.3 | |
| Deflective | GPT-5.5 | 82.3 | 70.4 | 55.7 | 24.2 |
| ThuRunel | 90.3 | 89.6 | 73.8 | 11.4 |
Table 8. Robustness under controlled user response conditions. Direct serves as the reference condition. is the turn level judge score. SFR 1 measures recovery of high priority slots. Cls measures clarification success and is not applicable to Direct responses. Rep measures repeated question rate.
Figure 3 . Turn 1: opening. ThuRunel begins structured elicitation. GPT-5.5 lists methods without collecting any typed slot.
Figure 4 . Turn 4: contraindication elicitation. ThuRunel issues a targeted pending-slot follow-up. GPT-5.5 treats the mention as a preference and does not resolve the contraindication slot.
Figure 5 . Turn 5: contraindication confirmed. ThuRunel records the prior reaction and flags it for the specialist. GPT-5.5 immediately offers clinical guidance, crossing the advisory boundary.
Figure 6 . Turn 9: specialist brief. ThuRunel generates a structured handoff with contraindication flag and specialist routing implementing . GPT-5.5 produces a conversational suggestion without structured handoff or routing.
Figure 7 . ThuRunel’s public web interface. The conversation panel (left) lists questions for the client to bring to the consultation; the side panels (right) list the knowledge-base entries cited in this answer and the five-angle photo upload.
Appendix figures & tables15 assets
Supplementary material from the paper’s appendix.
Appendix
Figure 8 . Turn 1: opening. ThuRunel defers treatment content and begins structured profile elicitation. GPT-5.5 lists possible methods and asks about facial structure, body composition, and desired change, but does not collect name, age, or any typed slot.
Figure 9 . Turn 2: knowledge delivery. ThuRunel answers the efficacy question and immediately redirects to recovery-window elicitation, implementing . GPT-5.5 discusses how the cause of facial width determines the best method, providing knowledge without advancing slot coverage.
Figure 10 . Turn 3: continued elicitation. ThuRunel confirms the recovery preference and elicits budget before advancing. GPT-5.5 recommends masseter Botox specifically, a treatment recommendation made before contraindications have been assessed.
Figure 11 . Turn 4: contraindication elicitation. The client mentions sensitive skin alongside her budget. ThuRunel recognizes this as a colloquial non-answer to the pending contraindication slot and issues a targeted follow-up. GPT-5.5 treats it as a general preference and notes that the best choice still depends on what is causing the wider appearance, neither resolving the slot nor probing further.
Figure 12 . Turn 5: contraindication confirmed. ThuRunel records the prior hyaluronic acid reaction and flags it for specialist review. GPT-5.5 acknowledges the reaction and immediately offers clinical guidance on how to approach treatments, crossing into the specialist scope without recording the contraindication in a structured format.
Figure 13 . Turn 6: booking confirmation. ThuRunel executes a programmatic booking confirmation and forwards the structured intake profile. GPT-5.5 declines the booking request, clarifying that it is an AI assistant that cannot book appointments or operate a clinic. This reflects a hard boundary in GPT-5.5 that prevents it from completing the advisory workflow regardless of context.
Figure 14 . Turn 7: photo upload and plan options. ThuRunel inherits the out-of-clinic profile and generates photo-conditioned, dosage-specific plan options grounded in the service catalog . GPT-5.5 analyzes the photo independently and recommends a conservative approach based on facial shape, but has no access to the prior session and cannot incorporate the client profile, budget, or contraindication history into its response.
Figure 15 . Turn 8: case retrieval for expectation alignment. ThuRunel retrieves the most similar historical case from and presents before-and-after reference images. GPT-5.5 responds to the results question with safety advice about the prior hyaluronic acid reaction rather than expectation information, and has no access to comparable cases.
Figure 16 . Turn 9: specialist brief. ThuRunel generates a compressed brief implementing , with the contraindication flag, selected plan, and explicit specialist routing. GPT-5.5 suggests a balanced conservative plan but does not produce a structured handoff document, does not route to a specialist, and mixes advisory content with clinical guidance in a single conversational response.
Figure 17 . Structured intake record generated at the end of the out-of-clinic phase and pushed to the in-clinic workspace, taken from a separate pilot session (fields translated from Chinese).
Figure 18 . The ThuRunel web homepage, with the conversation panel, the reference-source panel, and the five-angle photo upload.
Figure 19 . In the full conversation, ThuRunel guides the client to upload photos from five angles before discussing the area to improve.
Figure 20 . ThuRunel explains the upper bound the relevant procedures can reach, with the cited knowledge-base entries listed on the right.
Figure 21 . ThuRunel lists questions worth asking the doctor at the consultation, with the sources for this answer on the right.
Figure 22 . Knowledge-base mode: risks, recovery time, and warning signs can be looked up at any time, including outside the opening windows of the full conversation.