Tom Jones

Bootstrapped solo founder. Building Tīrtha, a cost efficient local AI engine with memory and verification for reliable long horizon agents.

Rock Hill Sc. Joined on Jun 27, 2026 https://tirtha.ai/

Work

Founder and CEO of Tirtha.ai

Tom Jones

Oct 11

We asked a decision model to check our AI's work. It got good the day we stopped asking it to think.

#ai #llm #testing #machinelearning

6 min read

Want to connect with Tom Jones?

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

Create Account

Already have an account? Sign in

Tom Jones

Oct 6

Our AI judge listens to how sure you sound. Our first number about it was wrong.

#ai #testing #machinelearning #llm

6 min read

How simple code hides past design choices

Tom Jones

Sep 27

Implementation is where judgements go to become invisible

#discuss #testing #programming #ai

10 min read

Tom Jones

Sep 25

Crystals: agent memory that arrives before you act

#ai #agents #architecture #llm

10 min read

Tom Jones

Sep 25

I shipped the same bug twice with the warning sitting one character-cap away

#ai #agents #architecture #llm

7 min read

Tom Jones

Sep 25

Escalating to the better model made 34 answers worse

#ai #llm #architecture #benchmarking

12 min read

Tom Jones

Sep 25

An index keyed to the diagnosis is a glossary

#discuss #ai #testing

6 min read

Tom Jones

Sep 23

What Grok catches, what Codex catches, and what the pair costs us

#ai #llm #testing #programming

8 min read

Tom Jones

Sep 23

Your agent already has the answer. You are paying it to look again.

#discuss #ai #rag #architecture

13 min read

Tom Jones

Sep 18

Your hooks are a fence. They could be a body.

#showdev #ai #agents #devtools

10 min read

Tom Jones

Sep 18

Whole notes, not fragments: the retrieval half

#ai #memory #rag #devtools

7 min read

Tom Jones

Sep 17

Crystal memory: notes that arrive when you act, not when you go looking

#showdev #ai #python #opensource

18 min read

Tom Jones

Aug 18

Who your model works with matters more than which model you picked

#ai #llm #testing #programming

23 min read

Tom Jones

Aug 12

The agent that remembers, and never decides

#discuss #ai #architecture #productivity

10 min read

Tom Jones

Aug 9

Microsoft measured our thesis, and we still cannot quote ours

#ai #testing #benchmarking #programming

7 min read

Tom Jones

Aug 7

A stable aggregate is not a stable measurement

#benchmarking #testing #ai #datascience

4 min read

Tom Jones

Aug 7

We stopped choosing models and started choosing what to measure

#ai #llm #benchmarking #engineering

8 min read

Tom Jones

Aug 2

I built a page that only publishes numbers surviving our own checks. It caught me inside the hour.

#showdev #ai #testing #webdev

5 min read

Tom Jones

Jul 29

We built a router to predict when a cheap model is enough. It does not work.

#ai #machinelearning #llm #devops

4 min read

Tom Jones

Jul 26

Tokens Are Not the Unit

#ai #llm #programming #architecture

12 min read

Tom Jones

Jul 26

Your company already runs the hardest parts of an AI agent platform

#ai #llm #architecture #productivity

9 min read

Tom Jones

Jul 5

Everyone is hardening the structure. Nobody is passing down the why.

#ai #llm #machinelearning #agents

3 min read

Tom Jones

Jul 4

I didn't mean to build this. Looking for testers anyway.

#ai #webdev #opensource #startup

2 min read

Tom Jones

Jul 2

I got my coding agent to tie the frontier for about 8x less. Here is the honest benchmark.

#showdev #ai #machinelearning #llm

3 min read

Tom Jones

Jun 29

Serving cheap when two models agree: a measured cost lever

#testdev #ai #llm #infrastructure

5 min read

Tom Jones

Jun 28

Frontier-Quality Coding at Cheap-Tier Cost: What We Built, and How We Measured It

#ai #testing #llm #devops

5 min read

Tom Jones

Jun 28

The Two-Channel Problem: Structure and Soul for Reliable Long-Horizon Agents

#ai #llm #machinelearning #agents

4 min read

DEV Community

Tom Jones

Badges

Community Favorite 2

8 Week Community Wellness Streak

4 Week Community Wellness Streak

2 Week Community Wellness Streak

1 Week Community Wellness Streak

Writing Debut

We asked a decision model to check our AI's work. It got good the day we stopped asking it to think.

Want to connect with Tom Jones?

Our AI judge listens to how sure you sound. Our first number about it was wrong.

Implementation is where judgements go to become invisible

Crystals: agent memory that arrives before you act

I shipped the same bug twice with the warning sitting one character-cap away

Escalating to the better model made 34 answers worse

An index keyed to the diagnosis is a glossary

What Grok catches, what Codex catches, and what the pair costs us

Your agent already has the answer. You are paying it to look again.

Your hooks are a fence. They could be a body.

Whole notes, not fragments: the retrieval half

Crystal memory: notes that arrive when you act, not when you go looking

Who your model works with matters more than which model you picked

The agent that remembers, and never decides

Microsoft measured our thesis, and we still cannot quote ours

A stable aggregate is not a stable measurement

We stopped choosing models and started choosing what to measure

I built a page that only publishes numbers surviving our own checks. It caught me inside the hour.

We built a router to predict when a cheap model is enough. It does not work.

Tokens Are Not the Unit

Your company already runs the hardest parts of an AI agent platform

Everyone is hardening the structure. Nobody is passing down the why.

I didn't mean to build this. Looking for testers anyway.

I got my coding agent to tie the frontier for about 8x less. Here is the honest benchmark.

Serving cheap when two models agree: a measured cost lever

Frontier-Quality Coding at Cheap-Tier Cost: What We Built, and How We Measured It

The Two-Channel Problem: Structure and Soul for Reliable Long-Horizon Agents