Continue from our previous task?
✧I can review context and propose the next step.
+ASK JEONGSONGI↗
Conversational AI
Interactive dialogue and context experiments
JEONGSONGI / MODEL-AGNOSTIC AGENT R&D
JEONGSONGI.
An AI core for conversation, code and real workflows.
Our core agent platform researches memory, retrieval, tool-use, independent evaluation and vision integration, building on documented 27B and LoRA STEP160 experiments.
Research in progress / not a production-ready or qualified beta
GPU research roadmapJEONGSONGI
CORE AI R&D
CORE AI R&D
PERCEPTIONLanguage · Vision
MEMORYMemory · Context
REASONINGReasoning
TOOLSGated Tools
APPROVALHuman Approval
EVALUATIONIndependent Evaluation
27B · LoRA STEP160 · MODEL-AGNOSTIC AGENT RESEARCH
WHAT THE PLATFORM CAN DO
A reusable AI core for far more than one market.
See the building blocks of conversation, coding assistance, domain learning, memory and safely gated tools.
JEONGSONGIMODEL-AGNOSTIC AI R&D
01def check_order(order):
02 sku = resolve(order.item)
+ require_approval(sku)
+ run_tests()
DIFF VIEW +2 CHANGES
Coding agent R&D
Code analysis, change proposals and testing under development
Field feedback→Train & test→New candidate
LoRASTEP 160
FEEDBACKHOLDOUT
Domain language adaptation
LoRA STEP160 research recorded; evaluation continues
Propose→Review→Approve & run
🛡
HUMAN APPROVAL REQUIREDREAD → PROPOSE → CONFIRMTools with human approval
Proposal, review, approval and controlled execution
27B27B model R&D
LoRA 160160 LoRA steps
VRAM / OOSLong context / VRAM
HUMAN GATEIndependent evaluation
NVIDIA / GPU RESEARCH
Real customers and working prototypes. Compute is the next bottleneck.
Reliable conversation, coding workflows, learning and tools require repeated long-context, concurrent and multimodal evaluation on adequate GPU infrastructure.
INFERENCEFINE-TUNINGPARALLEL EVALMULTIMODAL
01 / EVIDENCE-BASED DEVELOPMENTClear evidence.
Clear evidence.
Clear next milestones.
27B / LoRA experiments
STEP160 training and component-level tests are recorded.
Reliability and evaluation
GPU memory, long context, tool use and independent evaluation remain in progress.
Vision and operational tools
We aim to connect EYEBOX, inventory and approved business actions.
Iterative evaluation is strengthening reliability, tool use and field-facing functionality across the research roadmap.
02 / DESIGN PRINCIPLESModels may change.
Models may change.
Accountable workflows endure.
CONTEXT
Project continuity, grounded memory and retrieval
TOOLS
Authorized specialist tools with clear system ownership
HUMAN APPROVAL
High-impact changes remain proposals until human confirmation
INDEPENDENT EVAL
Separate quality, runtime, recovery and out-of-sample evidence
27B / LoRA R&D RECORD
FROM THE MODEL TO THE SYSTEMBeyond the model
A good answer is not the same as
a reliable business action.
The research goes beyond text generation: persistent context, provenance-aware memory, approval-gated tools, recovery and independent evaluation are necessary before any operational deployment.
- 27B and LoRA STEP160: documented experiments
- Long context, GPU memory and runtime reliability: open tests
- Multimodal, Coding Agent and EYEBOX integration: gated roadmap
HOW THE SYSTEMS CONNECTDistinct roles, one direction
EYEBOX sees the commodity.
Jeongsongi connects the context.
EYEBOX
Capture, visual evidence and expert quality criteria
JEONGSONGI
Memory, reasoning, gated tools and evaluation
QUICKINVENTORY
Human-approved purchase, inventory and settlement
We are linking product capabilities and Jeongsongi AI in stages, with appropriate access controls and human approval for consequential operations.
03 / WHY COMPUTE MATTERSGPU capacity enables
GPU capacity enables
rigorous experiments, not hype.
We seek research and development compute to validate 27B long-context memory, reliable tooling, GPU recovery, independent evaluations and specialist EYEBOX vision models.
Discuss GPU collaborationRESEARCH QUEUE 0127B / LoRA / EVALUATION
RESEARCH QUEUE 02VISION / QUALITY / LABELS
RESEARCH QUEUE 03TOOLS / HUMAN GATES
RESEARCH QUEUE 04NEW LOTS / OUT-OF-SAMPLE
ACTIVE AI RESEARCH · EVALUATION ROADMAP
MULTI-INDUSTRY VISION / MODULAR AGENT CORE
One core. Many potential industries.
Raw-material trading is our real-world testbed—not the boundary of the AI architecture.
Cross-industry applications are the development roadmap, not completed commercial deployments.