Solution
Flow-driven, object-centric custom agents
When work is not a single conversation but a long-lived business object moving through stages, channels, and systems until a defensible terminal state — you need reliable long-running workflows plus agentic judgment, anchored in one case ledger.
Stages & events
Explicit flow — not a one-off script
Calendar truth
Waits & SLAs survive restarts
Case ledger
One record for proof & audit
Governed scale
Conversation quality (IQA), compliance, supervisor review, spend governance
Chat projects vs. case programs
A re-KYC packet is not a message thread. A vehicle hypothecation file is not a ticket. These are cases with identity, SLAs in calendar days, and auditable exits.
Legacy — chat-first
Session amnesia
State in a browser or transcript — repeated asks and compliance risk.
Tool sprawl
Integrations fire ad hoc — actions contradict CRM / LMS truth.
No calendar truth
Follow-ups are manual — SLAs and regulator-facing timelines slip.
Weak exit semantics
“Resolved” is a label, not proof — audit suffers.
Sentient — object-first
Object-led
One case ledger ties conversations, documents, tools, and approvals.
Event-fed
Typed signals drive transitions — composable suites, not forked code.
Policy-bound
Agent judgment within guardrails; deterministic actions and durable waits.
Provable exits
Terminal outcomes you can explain — versioned policy and artifact linkage.
Three definitions — keep them crisp
Product marketing and architects align when flow, persistence, and object mean the same thing in every deck.
Flow-driven
Explicit stage model: allowed states, transitions on typed events (customer replied, document ingested, webhook received), and wait strategies — sleep until date, wait for signal. A compact workflow layer; branching policy via agent decisions plus guardrails.
Persistent
Durability across time and failure: regulatory cooling-off, dealer upload deadlines, and callback promises survive process restarts — state belongs to the workflow layer and stores, not an LLM session.
Object-centric
The case (loan file, policy, shipment, employee packet) is the unit of truth for “done.” Conversations are projections of the case; memory binds to case graphs for auditable coherence across channels.
Continuous control loop
Object-led, event-fed, policy-bound
Replace “one more AI assistant” with a loop around a single case object: signals in → context from object and history → agent decision → guardrails → deterministic actions → durable wait → repeat until a meaningful terminal outcome.
01
Signals in
Webhooks, replies, docs
02
Context
Object + history
03
Agent decision
Judgment + guardrails
04
Actions
Channels & tools
05
Durable wait
Timers, SLAs, signals
06
Terminal / loop
Proof or next stage
Schematic — brain vs timing layer (conceptual)
Cycle time, stage conversion, cost-to-complete, and conversation quality (IQA) — not raw message counts
Six solution tiles — same engine, different objects
Cross-industry patterns for decks and proposals. The nouns change — the mechanics stay aligned.
Regulatory custody & hypothecation
Case spine
Secured lending file — loan + collateral + regulatory artifacts
Why persistence
Mandates have calendar clocks; parties do not move at chat speed
Example exits
RC endorsed / charge registered; exceptions escalated with evidence
Periodic re-identification (re-KYC)
Case spine
KYC packet — risk tier, docs, attestations
Why persistence
Time-boxed refresh; stop/limit until complete
Example exits
Completed / exempted with reason; audit trail for supervisors
Clinical access & authorization
Case spine
Authorization case — clinical + administrative fit
Why persistence
Peer-to-peer time; multi-party document chase
Example exits
Approved / denied / appealed with artifact linkage
Claims evidence & subrogation intake
Case spine
Claim object — coverage, liability, third parties
Why persistence
FNOL is day zero; truth emerges over weeks
Example exits
Settled / denied / litigation path with document completeness
Order-to-cash exception desk
Case spine
Order / invoice dispute object
Why persistence
Exceptions stall cash; nudges respect credit policy
Example exits
Matched / credited / legal review with ERP sync
Regulated fulfillment & field proof
Case spine
Service order — install, swap, compliance proof
Why persistence
Field SLAs and safety constraints
Example exits
Closed with proof; failed visit with reschedule + reason
Platform in conversation
9 domains, one platform
KYC, loan onboarding, insurance renewal, vendor compliance, prior auth — all object-centric.
Aria · WhatsApp
Re-KYC: periodic identity refresh, banking — Meenakshi's KYC documents expired. The agent reaches out 14 days before account restriction under RBI mandate, walks her through document submission on WhatsApp, catches a blurry Aadhaar back, and completes the refresh in one session.
Hover to pause · click any domain above · auto-advances through all 9
Where programs earn their keep
High-consequence workflows need conversation quality scorecards (IQA), compliance templates, and structured supervisor review — not a failure mode.
Re-KYC & stop-the-line governance
Risk alignment under board and regulator scrutiny: periodic refresh, event-triggered refresh, tiered evidence, and hard stops until the packet is complete. The KYC packet object carries party anchors, risk tier, evidence checklist per item, and exception reasons — agents negotiate how; the workflow layer enforces when reminders fire and when accounts restrict.
Auto loan RC hypothecation
Collateral perfection as a lifecycle: milestones from disbursement to charge registration, with artifacts and multi-party timers. Agents explain timelines and proof; the workflow layer does not lose deadlines. Terminal states are machine-checkable where possible and human-attested where not — the ledger answers what was true, when, under which policy version.
Prior authorization — multi-party clock
Clinical validity, benefits, network, and peer review each start and stop clocks. The case persists across uploads, payer responses, P2P scheduling, and appeals. The agent coordinates communication without replacing medical judgment — persistent alignment to reduce avoidable denials and keep SLAs visible.
Operating the program — personas & metrics
| Persona | What they need | How the platform supports it |
|---|---|---|
| Policy owner | Defensible rules and versioning | Published templates, version history, pilot vs production (Test / Live) |
| Ops lead | Throughput without chaos | Exception-first consoles from workflow signals |
| Specialist | Summary-first decisions | Structured escalations with policy context |
| Auditor / risk | Proof, not screenshots | Flags, evaluations, artifact linkage to cases |
| Platform admin | Healthy integrations | Event registry, routing health, observability |
Metrics that matter combine cycle time, stage conversion, cost-to-complete (spend governance / FinOps), and quality (conversation quality / IQA).
The unheroic pillars that make “custom” controlled
Spend governance
Long-running programs burn voice, messaging, and model tokens. Budgets and virtual keys keep automation profitable and attributable.
Compliance
Policy evaluation as software — templates with severities and record-level flags, not PDF theater.
Durability
Calendar time is the domain. The workflow layer owns sleep, wake, and reconciliation when reality changes.
Complex scenarios — plain explanations
Wrong document uploaded
Reject with reasons, requeue checklist, preserve attempts for audit
Third party delays (e.g. dealer)
Timers + evidence of nudges; escalation with neutral facts
Regulator asks what happened in March
Versioned policies + case history + flags on touches
VIP / vulnerable customer
Route to specialized suites with tighter guardrails and supervisor review
Model or vendor outage
Deterministic fallbacks; the workflow layer does not forget the case
White papers & sneak-peek demos
Dedicated narratives for financial services and non-financial buyers — flow-driven, object-centric custom agents on the Thuriyam / Sentient Agents platform. Add PDFs under public/whitepapers/ when ready.
Flow-driven, object-centric custom agents
Secured lending, re-KYC, capital-markets-facing proof, and regulated fulfillment — calendar clocks, artifact linkage, and supervisory-ready narratives.
Download PDFPlace file at public/whitepapers/flow-driven-object-centric-custom-agents-financial.pdf
Custom case programs across industries
Healthcare authorization, P&C claims intake, O2C exceptions, and field service — same object ledger, flow, and governance story without core-banking framing.
Download PDFPlace file at public/whitepapers/flow-driven-object-centric-custom-agents-general.pdf
Sneak-peek demo videos
Placeholder players — embed YouTube, Vimeo, or Mux in production.
Custom program lifecycle
Stage model, typed events, and waits — from intake to defensible terminal state.
Request full demo reel →Case ledger & artifact graph
One object ties conversations, documents, tool calls, and approvals for audit.
Request full demo reel →Governance console
Conversation quality (IQA), compliance templates, supervisor review queues, and program budgets in one operating picture.
Request full demo reel →Obligations automated — without bespoke code or fragile chat UI
Flow-driven, persistent, object-centric custom agents bind judgment to reliable long-running workflows and truth to a case ledger — with conversation quality (IQA), compliance, supervisor review, spend governance, and auditability so custom still means controlled.