Application

Fusing Reasoning Models, Generative AI & LLMs: The Winning Formula

Why This Trio Excels:

The models

By orchestrating three complementary AI paradigms—reasoning engines, generative networks, and large language models—we unlock systems that are both creatively fluent and rigorously reliable.

Example Architecture:
LLM (e.g., GPT):
Handles natural language tasks like understanding queries, generating responses, summarizing data.

Reasoning Engine:
Takes over for tasks that need logic, such as math, planning steps, verifying facts.

Agentic Framework (e.g., AutoGPT, LangChain, OpenAI Function Calling):
Coordinates everything. It can:

Break a goal into subtasks:

  • Decide when to ask the LLM for help
  • Trigger the reasoning module when logic is required
  • Store results in memory or a database
  • Adapt behaviour over time

Scenario: AI Agent Solving a Research Task

Let’s walk through an example workflow of how an Agentic AI might operate – deciding when to use Large Language Model (LLM) for text understanding/generation, Reasoning Model for logic-heavy tasks and external tools/APIs for specialized functions like search or math.

Summarize the latest research on Alzheimer’s treatments, check which ones are FDA-approved, and recommend the most promising one with scientific reasoning.

Step1 – Understand the Task

LLM is used to parse the request, recognize subtasks:

  • Get latest Alzheimer’s research
  • Identify FDA-approved treatments
  • Analyze & recommend best one
  • LLM used for task decomposition
Step 2: Search for Information

The agent uses a web search tool/API to gather recent publications or clinical trial data.
External tool used for data retrieval.

Step 3: Summarize the Findings

LLM is prompted to generate a summary of the top research papers.
LLM used for summarization & synthesis.

Step 4: Verify FDA Approval

The agent uses an FDA database API, or structured data source. A Reasoning Model or logic component compares drug names against approval status.
Reasoning Model + Tool used for fact-checking.

Step 5: Recommend Best Treatment

LLM proposes a candidate.
Reasoning Model performs a logical evaluation efficacy, trial phases, side effects, approval status.
The agent ensures the recommendation is based on evidence, not just text fluency.
LLM + Reasoning Model collaborate on decision-making.

Step 6: Generate the Final Answer

LLM crafts a clear, professional response in natural language, citing sources and reasoning steps.
LLM used for final output formatting.

Why This Hybrid Setup Works

Summary

Type
Key Role
Strengths
Weaknesses
Reasoning Models
Logical inference and problem-solving
Accuracy, consistency
Limited generalization, slow
LLMs / Generative AI
Natural language generation and understanding
Versatile, broad, creative
Can hallucinate, lacks deep reasoning
Agentic AI
Goal-directed, autonomous action
Agentic AIIndependence, planning, coordination
Still experimental, hard to align and control

Flexibility

Fully programmable

Algorithm agnostic

Host processor agnostic

RISC-V cores to offload host
& run AI completely on-chip.

Memory

Capacity

HBM: 288GB

Throughput

HBM: 8 TB/s

Performance

Tensor core (dense)

FP16: 800 Tflops
FP8: 3200 Tflops

General Purpose

FP32: 25 Tflops
FP16: 50 Tflops
FP8: 100 Tflops
Close to theory efficiency

Explore Jotunn 8

Introducing the World’s Most Efficient AI Inference Chip.