Developer Builds AI Agent 'Milo' to Power an Interactive, Data-Driven Portfolio

·2 views

A developer has created Milo, an AI agent layered on top of a personal developer portfolio to make it interactive and programmable. The system uses PostgreSQL as its central data store, a FastAPI backend for controlled data access, and a React frontend, with Milo serving as the conversational interface. Rather than granting the AI direct database access, Milo operates through backend-defined tools, keeping the reasoning layer separate from application data and business logic. The agent is also integrated with Telegram, enabling the portfolio owner to manage content remotely, with sensitive actions requiring explicit confirmation. A notable feature is CV curation, where Milo can analyze a job description and generate a tailored CV using only verified portfolio data, without fabricating qualifications.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Developer Builds Production-Grade AI Telegram Motivation Bot with Node.js and Gemini

A developer has published a detailed architectural breakdown of an AI-powered Telegram motivation bot built using Node.js, Google Gemini, and the Telegram Bot API. The bot goes beyond simple prompt-response interactions by incorporating features such as scheduled messages, user personalization, streak tracking, feedback collection, and telemetry. To ensure reliability, the system includes retry logic, a model priority fallback chain, and an in-memory cache to reduce redundant API calls. Multiple persona archetypes and content modalities — such as Stoic, Warrior, and Pragmatic — are used to generate varied motivational messages without requiring major architectural changes. The project, shared on DEV Community, is structured with modular components covering bot commands, services, and data management, and is tested using Jest.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How to Diagnose Missed MT5 EA Entries Without Creating False Signals

Developers building Expert Advisors in MetaTrader 5 sometimes encounter missed trade entries that appear valid on the chart but are silently blocked by internal filters. A structured diagnostic approach recommends logging the full symbol, timeframe, bar timestamp, and the specific decision point that halted entry progression, rather than relying on a single boolean flag. Collapsing multiple outcomes into one variable obscures whether the failure was due to unavailable data, an unqualified signal, or a gate condition such as an outside-session block. A JavaScript fixture demonstrates a four-stage classification model — input, signal, gate, and request — showing that data readiness must be verified before any signal value is trusted. The guide emphasizes that reaching the request stage does not confirm a broker accepted an order, and that MT5's sequential tick processing means not every market tick is guaranteed to become a distinct handler call.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer Builds AI Italian Copy Reviewer That Cites Its Own Style Guide

An Italian developer working on software website localization built an AI-powered copy review agent to catch recurring translation errors in Italian marketing text. The tool, submitted for the Sanity Challenge, reads a team-authored Italian style guide stored in a Sanity Knowledge Base containing 15 rules and 19 glossary terms drawn from real shipped product copy. When given a string to review, the agent returns a verdict, a corrected version, and the specific style guide entries it relied on to reach its conclusion. The Knowledge Base was built from two sources — a custom production dataset and Microsoft's public Italian Localization Style Guide — with conflicts between them flagged and manually resolved before final testing. In benchmark tests on 12 held-out string pairs, the grounded agent matched the performance of the same model running without the Knowledge Base, though accuracy gains from the structured guide are still being measured.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

AI Tools Are Helping Diners Decode Menus and Track Nutrition Instantly

Artificial intelligence is being applied to restaurant menus to help diners make more informed, personalized food choices. Using natural language processing and machine learning, AI systems can parse menu text, identify ingredients, and estimate nutritional values by cross-referencing standardized databases such as the USDA FoodData Central API. The technology addresses long-standing pain points for people managing allergies, calorie goals, or specific dietary needs, who previously had to rely on guesswork or incomplete menu information. Restaurant operators also stand to benefit, as AI can reduce the time and cost involved in manually updating nutrition details and accommodating evolving dietary trends. Developers are already building practical implementations that map dish ingredients to nutrient profiles and generate tailored meal recommendations for individual users.

0 comments Read more at DEV Community