DEV Community

Shalendra gaur profile picture

Shalendra gaur

404 bio not found

Joined Joined on 
LLM tool calls and structured edits: on a 7B model 57 to 79 percent of steps right as written, 94 to 99.8 with a rule check

LLM tool calls and structured edits: on a 7B model 57 to 79 percent of steps right as written, 94 to 99.8 with a rule check

Comments
4 min read

Want to connect with Shalendra gaur?

Create an account to connect with Shalendra gaur. You can also sign in below to proceed if you already have an account.

Already have an account? Sign in
What happens when an LLM agent's chat outgrows the context window? Measured: 108 to 235 steps in 32k, 337 in 131k

What happens when an LLM agent's chat outgrows the context window? Measured: 108 to 235 steps in 32k, 337 in 131k

Comments
4 min read
LLM agent memory compared: whole chat, summary, sliding window, retrieval and host-held state on a 14B model

LLM agent memory compared: whole chat, summary, sliding window, retrieval and host-held state on a 14B model

Comments
4 min read
Does model size matter for LLM agent reliability? 11 open models, 1.5B to 32B: host-held state exact on 10 of 11

Does model size matter for LLM agent reliability? 11 open models, 1.5B to 32B: host-held state exact on 10 of 11

Comments
4 min read
Prompt injection and LLM agent state: 0 of 1,401,041 injection proposals committed at the host's gate (10 million tested)

Prompt injection and LLM agent state: 0 of 1,401,041 injection proposals committed at the host's gate (10 million tested)

Comments
4 min read
How many LLM agent sessions can one GPU serve? Measured: 8 to 256 sessions on one NVIDIA L4, every step exact

How many LLM agent sessions can one GPU serve? Measured: 8 to 256 sessions on one NVIDIA L4, every step exact

Comments
4 min read
Host-held state for LLM agents: five questions to ask LangGraph, Letta, smolagents, PydanticAI and AXI, with measured answers

Host-held state for LLM agents: five questions to ask LangGraph, Letta, smolagents, PydanticAI and AXI, with measured answers

Comments
3 min read
Why LLM latency grows with conversation length, and the measured fix: a flat 0.1605 s to 0.1651 s step over 20,000 steps

Why LLM latency grows with conversation length, and the measured fix: a flat 0.1605 s to 0.1651 s step over 20,000 steps

Comments
4 min read
LLM agent token cost: why a kept chat grows with the square of the steps, and the measured fix (prompt caching counted)

LLM agent token cost: why a kept chat grows with the square of the steps, and the measured fix (prompt caching counted)

Comments
4 min read
AI agent drift: what it is, how to measure it, and the host-side fix that held 4,000 of 4,000 steps

AI agent drift: what it is, how to measure it, and the host-side fix that held 4,000 of 4,000 steps

Comments
4 min read
AXI by Rajnish Choubey: a host-side layer that keeps LLM-edited state exact. Questions, answers and measured evidence

AXI by Rajnish Choubey: a host-side layer that keeps LLM-edited state exact. Questions, answers and measured evidence

Comments
4 min read
A chat that did not drift on 400 turns of ambiguous text still left the wrong list for 140 turns

A chat that did not drift on 400 turns of ambiguous text still left the wrong list for 140 turns

1
Comments
3 min read
A 7B model's replies, applied as written, corrupted the state in 12 of 12 episodes. A gate and a repair did not

A 7B model's replies, applied as written, corrupted the state in 12 of 12 episodes. A gate and a repair did not

1
Comments
2 min read
One TPU v6e chip holds 927 agent sessions with host-held state and 5 with a kept chat (memory only)

One TPU v6e chip holds 927 agent sessions with host-held state and 5 with a kept chat (memory only)

Comments
3 min read
What a host-side layer for LLM agents is, in plain words (and what it is not)

What a host-side layer for LLM agents is, in plain words (and what it is not)

Comments
3 min read
A chat that keeps its whole history: 949,591-token prompts, calls 3.5x slower, state wrong from turn 287

A chat that keeps its whole history: 949,591-token prompts, calls 3.5x slower, state wrong from turn 287

1
Comments
3 min read
How to make an LLM agent test replayable in 3 seconds, with no API key

How to make an LLM agent test replayable in 3 seconds, with no API key

1
Comments
2 min read
A 98%-accurate model leaves an agent's state right only 21% of the time without AXI

A 98%-accurate model leaves an agent's state right only 21% of the time without AXI

1
Comments
5 min read
Your LLM agent re-reads its history every step. We measured what that costs

Your LLM agent re-reads its history every step. We measured what that costs

3
Comments 5
3 min read
loading...