20 posts tagged with "AI and Machine Learning"

Quick catch-up between Johnny (@jnxmas) and Victor (@zcbtvag) covering Opik integration progress and OCR pipeline challenges.

Summary​

Johnny showcased recent progress on Opik prompt optimization. A new architecture is in place where prompts are no longer hardcoded but managed via an Opik Prompt Library. The "Charter Validation" prompt has already been optimized using this system.

A new mock-up feature can automatically generate contributions (even with violations) from existing meeting reports. The goal is to create a robust dataset to test and improve the validation agent. However, this auto-generation currently produces repetitive content - a challenge that will need addressing by identifying and aggregating duplicate contributions.

Today we completed a major architectural milestone: modular prompt management for Forseti461. Each feature now has its own versioned prompt in Opik, enabling independent optimization and A/B testing.

Forseti461 is an AI agent that automatically moderates citizen contributions to participatory democracy platforms — approving only concrete, constructive, locally relevant ideas while rejecting personal attacks, spam, off-topic posts, or misinformation, and always explaining decisions with respectful, actionable feedback.

What if we could protect civic discourse at scale?

Goal: Create a systematic approach to test and improve our AI-powered charter validation system.

For the Encode Hackathon first submission, we focused on building the infrastructure to ensure Forseti 461 (our charter validation agent) catches all violations reliably. The key insight: you can't improve what you can't measure.

Here is the assessment of the catch-up call between @jnxmas and Victor regarding the Ò Capistaine project status and immediate priorities.

Summary of the Call​

@jnxmas and Victor discussed the immediate roadmap for the Opik/Commit to Change Hackathon MVP submission (deadline: ~1 day, 14 hours). Victor has successfully downloaded approximately 4,000 PDFs (including ~3,965 deliberation documents), though he noted some potential duplicates and that the download process was synchronous and could be optimized later. He has committed changes to a development branch but not yet merged them, preferring to use GitHub as a medium to exchange the code while keeping the large PDF dataset local (or shared via a specific sub-directory). The team agreed on a strategy for the Hackathon demo deployment. Instead of using Vercel, which complicates environment variable management for their specific security setup (ngrok, multiple API keys for Opik, Firecrawl, Gemini, etc.), @jnxmas will run the demo from his local machine using a secure, paid ngrok tunnel (ocapistaine.ngrok.app). This setup allows the jury to interact with the Streamlit UI (restricted to Ollama for the external demo) while the team can continue testing other models (Gemini) locally. The architecture involves Locki.io -> Vaettir Orchestration -> Local Machine (Ocapistaine Agent).

Context: Audierne 2026 Election Platform

This document outlines how we will leverage the Google Gemini ecosystem (AI Studio, Flash models, and Agentic workflows) to accelerate the development of the Locki project. By utilizing these tools, we aim to bridge the gap between human ideation and automated N8N workflows, specifically for the Commit to Change hackathon and the subsequent election period.

Strategy Update: Leveraging Gemini & Agent Workflows​

This document summarizes key insights from the recent team sync following Jean-Noël's attendance at the Google Gemini workshop. The focus is on accelerating the Locki / Ò Capistaine project by transitioning from manual human analysis to automated agent workflows.

This training consolidates the operational and technical foundations needed to run and execute agent/prompt optimization in team settings (e.g., hackathons and internal workshops).

It includes :

• eval-driven optimization of LLM agent prompts using measurable metrics and iterative loops,
• including meta-prompting, genetic/evolutionary methods, hierarchical/reflective optimizers (HRPO), few-shot Bayesian selection, and parameter tuning.

This lecture, led by Abby Morgan, an AI Research Engineer, introduces AI evaluation as a systematic feedback loop for transitioning prototypes to production-ready systems. It outlines the four key components of a useful evaluation: a target capability, a test set, a scoring method, and decision rules. The session differentiates between general benchmarks and specific product evaluations, emphasizing the need for observability in agent evaluation. It demonstrates using OPIK, an open-source tool, to track, debug, and evaluate LLM agents through features like traces, spans, 'LM as a judge', and regression testing datasets.

After months of development on our horse racing analysis platform, we're taking a bold step forward. In 2025, Locki Labs shifts focus from real-time data aggregation to something far more ambitious: Valkyria — a temporal prediction system designed for scientific rigor and reproducible results.