About Us Paragon International, Inc. is a U.S.-based manufacturer of commercial concession equipment and food service products. We receive purchase orders from customers such as Amazon, Home Depot, distributors, school systems, and other commercial customers. Orders arrive by email in many different formats, including PDFs, Word documents, Excel spreadsheets, scanned documents, and occasionally photographed purchase orders. We are looking for an experienced AI Automation Engineer to design and build a production-ready system that automates our entire order intake process. This is not a simple chatbot project. We need someone who has successfully built business automation systems that combine AI, OCR, document processing, APIs, and workflow automation. Project Overview The system will monitor one or more Gmail inboxes continuously and automatically process incoming emails and attachments. The workflow should: * Monitor Gmail 24/7 for new incoming emails. * Download all attachments automatically. * Read: * PDF files * Microsoft Word documents * Excel spreadsheets * Scanned PDFs * Image files (JPG, PNG, TIFF, etc.) * Photographs of purchase orders * Use OCR when required. * Use AI to determine whether the email is: * Purchase Order * Quote Request * Cancellation * Return/RMA * Customer Inquiry * Other * Identify the customer automatically. * Extract all order information into a standardized data structure. * Detect duplicate purchase orders. * Automatically print valid purchase orders to our network printer. * Save documents into organized folders. * Rename files using a consistent naming convention. * Move processed emails into Gmail folders/labels. * Generate logs for auditing and troubleshooting. ## Future Phases The initial project focuses on reliable document processing and printing. Additional phases may include: * Sage 100 ERP integration * Automatic sales order creation * Inventory verification * Customer acknowledgment emails * Shipping workflow automation * Dashboard and reporting * AI exception handling * Multi-location printing We are looking for a long-term development partner who can continue improving the system over time. ## Required Skills Please apply only if you have strong experience with most of the following: * OpenAI API / ChatGPT API * Gmail API * OCR technologies (Tesseract, Azure Document Intelligence, Google Vision, AWS Textract, or similar) * Intelligent Document Processing (IDP) * PDF parsing * Workflow automation * Python * REST APIs * Windows automation * Network printing * Error handling and logging * AI document classification Experience with the following is a significant advantage: * n8n * Microsoft Power Automate * Make.com * ERP integrations * Sage 100 * Purchase Order processing * Manufacturing or distribution businesses ## Deliverables The completed solution should: * Run continuously with minimal supervision. * Be reliable enough for production use. * Handle errors gracefully. * Be well documented. * Be easy for our staff to maintain. * Be scalable as our order volume grows. ## To Apply Please include: 1. A description of similar automation projects you have completed. 2. Which automation platform you recommend (Python, n8n, Power Automate, Make, or another solution) and why. 3. Examples of AI document processing or OCR projects you've built. 4. Your experience integrating with ERP systems. 5. Your estimated timeline. 6. Your hourly rate or fixed-price proposal. Please begin your proposal with the phrase: **"I have built AI document automation systems."** This helps us identify applicants who have carefully read the project description. We are looking for a long-term partner, not just someone to complete a single project. If this project is successful, additional work will include ERP integration, warehouse automation, customer service automation, purchasing automation, and AI-driven business process improvements.

We are looking for a hands-on Forward Deployed AI Engineer to help build practical AI systems This is not a pure backend role and not a strategy-only consulting role. You will work close to end users, understand how their workflows actually operate, and then build AI-enabled tools that solve specific business problems. The ideal person is a strong software engineer who is comfortable with ambiguity, can communicate clearly with non-technical stakeholders, and can take an AI prototype from idea to something reliable and usable. What you will do - Learn the business workflows, systems, data, and constraints. - Build AI applications using Claude or similar large language models. - Use the right mix of prompting, retrieval, tool use, agents, and workflow automation. - Own delivery from scoping through prototype, testing, hardening, and handoff. - Create evaluations to determine whether the system is accurate, reliable, and safe enough to use. - Translate between domain experts and technical implementation. - Work carefully with sensitive or regulated data. - Document what you build so it can be maintained and reused. What we are looking for - Strong Python engineering skills. - Hands-on experience building with LLMs, preferably Claude or the Anthropic API. - Experience with RAG, structured prompting, tool use, evaluation, or agentic workflows. - Ability to operate independently in a messy, ambiguous environment. - Strong communication skills with both technical and non-technical stakeholders. - Track record of shipping working software, not just demos. - Comfort working with real-world data, integrations, and imperfect requirements. Helpful but not required - Prior forward deployed engineering, solutions engineering, or technical consulting experience. - Experience building AI tools for enterprise customers. - Experience in regulated or sensitive-data environments. - Familiarity with validation, auditability, traceability, or compliance-oriented workflows.

Project Budget: $500 (Strictly Milestone-Based) CRITICAL REQUIREMENT BEFORE APPLYING: Payment for this project is strictly tied to real-world performance metrics. Milestone 1 requires a live stopwatch test on a mobile device showing sub-1-second cache loading on repeat lookups. If you do not have deep experience with high-speed local database architecture and caching models, do not apply. We measure deliverables with a stopwatch, not excuses. Project Overview & The Long-Term Vision: I am building Reseller Bro, a powerful mobile utility application designed for on-the-go resellers to instantly analyze product values, get a FLIP/SKIP verdict, and SAVE to a digital cart in seconds. This app is just the initial foundation—the engineer who successfully delivers this backend infrastructure will have the opportunity to partner with us long-term to build out our entire ecosystem, including advanced B2B inventory management tools and our wearable AR glasses workflow (Bro Lens). The front-end user interface and layout are already mostly complete. We are anchoring our backend data pipeline to a high-speed eBay API model that automatically calculates smart market estimates for other secondary platforms. We need an expert developer to clean up our database cache, implement a tier-condition formula, handle minor UI adjustments, and add high-energy audio/vibration triggers. The Core Tasks & Milestone Payment Structure: Milestone 1: Sub-1-Second Database Caching ($100 Escrow) The Issue: The previous build incorrectly forced live AI image recognition to run on every single scan, causing a 7–8 second delay even on repeat lookups. The Fix: You must implement a proper local database caching layer (e.g., SQLite). The first initial scan of an item can take up to 7 seconds to run the AI workflow and fetch the initial marketplace data. However, on any repeat scan, the app must skip the AI image recognition entirely, read a cached unique text identifier/key, and instantly pull the results from the internal database in under 1 second. Milestone 2: Data Engine & Percentage-Based Estimation Matrix ($200 Escrow) The Fix: Connect the backend cleanly with the official eBay Browse API using our developer keys. Target Marketplaces: The app displays valuation metrics for four core platforms: eBay, Depop, Grailed, and Poshmark. Condition Matrix & Platform Estimation Formula: Because eBay utilizes a wide variety of specific conditions across different categories, you will build an automated mapping and calculation formula. The app will pull raw condition data initially from the eBay API, group it cleanly into our 3-tier user system, and then use those baselines to instantly calculate the estimated market values for the other three platforms (Depop, Grailed, and Poshmark). New Tier: Dynamically maps all brand-new and pristine variations data directly, including: "New", "Excellent", "Excellent - Refurbished", "Open box", "New with box", "New with defects", and "New without box". Good Tier: Dynamically maps all standard pre-owned and quality-certified variations data directly, including: "Very Good", "Good", "Used", "Very Good - Refurbished", "Good - Refurbished", "Pre-Owned", and "Certified Pre-Owned". Poor Tier: Maps heavy-wear options directly, such as "For parts or not working" or "Fair". If a specific item category lacks a true "poor" marketplace data option, the engine must automatically fall back to calculate a custom percentage markdown (e.g., 40% less) relative to that item's "Good" tier baseline. The results from these three tiers will automatically calculate estimated market values for Depop, Grailed, and Poshmark using an internal background multiplier. If a user wants to check the exact live screen on those blocked platforms, tapping a platform tile will trigger a direct, one-tap deep link search into that specific app or web page. Smart Category Specifications (Vehicle & Electronics Handling): If the AI detects an image of a Vehicle (cars, trucks) or high-value Electronics, the app must dynamically generate a quick-spec form for the user to confirm/fill in (Vehicles: Make, Model, Year, Mileage; Electronics: Brand, Model, Capacity). This structured data must be fed directly into the pricing API for precise accuracy. Milestone 3: UI Redesigns, Audio/Haptics, Live Deployment & Final Handoff ($200 Escrow) UI Tweaks: Implement minor visual layout edits and updates to a few existing UI screens to align with this new calculation model. This includes ensuring tiles are accurately labeled as "Estimated Value," verifying the condition buttons display perfectly, adjusting the selling platform logos/designs, and making a few structural changes to the "Saved Items" page (I will go over the exact design changes with you directly). Haptics & Custom Audio Cues: Implement device vibration triggers (via Web Vibrations API) to pulse the device exactly when a verdict hits the screen. Integrate custom short audio sound bites that trigger instantly on the verdict display: a high-energy "YEAH!" (Lil Jon style) sound bite for a FLIP verdict, and a "HELL NO!" sound bite for a SKIP verdict. Live Deployment: Once the final features are fully approved, you will be responsible for successfully deploying the production build live onto our hosting account so the application is fully operational. Clean up the codebase and hand over the complete, finalized source repository. Requirements: Deep expertise in backend optimization, API data pipelines, and high-speed local database caching. Proficiency in mobile development frameworks, front-end audio integration, and Node.js. Strong communication skills. You will work with an existing repository and must provide clean, documented code. To Apply: Please reply explaining exactly how you will structure the local database cache so that a repeat scan completely bypasses the image-recognition API step to hit the sub-1-second mark. Copy and paste this right onto Upwork. It has every single feature, condition category, and protective barrier built in!

We're building a data-driven platform in the beauty/cosmetics space. We've already collected a very large volume of scraped product and brand data, and it's messy - inconsistent formats, duplicates, missing fields, mixed languages, and unreliable values. Your job is to turn that raw data into something clean and reliable, and to build the web application on top of it. This is not a quick gig. We're looking for someone to stay with us for roughly a year and grow with the product. What you'll be doing Building and maintaining a Next.js (App Router) front end and back end Designing and managing our Supabase setup (Postgres schema, RLS policies, auth, storage, edge functions) Cleaning, normalizing, and deduplicating large beauty datasets (this is a major part of the role) Building data pipelines to process and validate incoming scraped data Writing transformation/normalization logic (units, brand names, categories, ingredients, pricing, etc.) Setting up data-quality checks and monitoring Iterating on features with us as the product evolves You should have Strong production experience with Next.js and Supabase (please point to specific projects) Solid SQL/Postgres skills - not just ORMs Real experience cleaning and normalizing large, messy datasets (Python/pandas or similar is a plus) Comfort working with scraped data and all the inconsistencies it brings Good written English and reliable communication Ability to work independently and own your part of the product Nice to have Experience in e-commerce, catalog, or product-data projects Familiarity with data pipelines / ETL tooling Some ML/NLP experience for entity matching or text cleanup Budget We expect this to run around $4,000–$5,000 USD per month depending on experience and hours. This is a long-term commitment, so stability and quality matter more to us than the lowest rate. To apply, please include: Similar past work - links or short descriptions of Next.js + Supabase projects, and at least one data-cleaning/normalization project you've done. Your time zone and your typical available hours - we need to know there's reasonable overlap. Your expected monthly budget within the range above (or your rate if hourly). Applications that don't address all three points will not be considered. A short note on how you'd approach cleaning messy beauty product data is a big plus. We read every proposal carefully and respond quickly to strong candidates. Looking forward to working with you.

We are seeking an experienced freelancer to set up an Odoo dashboard for our business. The ideal candidate will have a strong understanding of Odoo and its capabilities, as well as experience in creating customized dashboards. The project involves integrating various modules and ensuring seamless functionality. If you have a passion for Odoo and a knack for creating user-friendly interfaces, we would love to hear from you!

AI Engineer (RAG & Agentic Workflows). *LLM RESPONSES AUTOMATICALLY AVOIDED* We have already launched a production generative AI product that utilizes a custom Retrieval-Augmented Generation (RAG) architecture. We are now expanding the platform to include CRM intelligence, workflow automation, and agentic AI capabilities. This is **not** a prompt engineering role. Seeking an engineer with deep experience building and deploying production AI systems that combine LLMs with multiple structured and unstructured data sources. You should be comfortable walking into an existing, complex codebase, understanding the current architecture, and improving it. Existing AI Architecture Our current AI architecture consists of: * OpenAI embeddings * Embeddings stored in MongoDB * MongoDB Atlas Vector Search for retrieval * Retrieval from both structured SQL data and unstructured document collections * Existing tool/function-calling architecture **Please do not apply if you have not previously built or maintained production RAG systems using embeddings and vector search.** Experience specifically with **OpenAI embeddings and MongoDB Atlas Vector Search** is highly preferred. CRM Intelligence Layer We are currently building a CRM platform and need the AI to reason over CRM records, including the other records are RAG currently retrieves. You will be responsible for designing and implementing the AI integration layer that enables the LLM to intelligently retrieve and reason over CRM data. This work includes: * Designing AI tools/functions that expose CRM data to the LLM. * Implementing backend tool handlers that retrieve CRM records. * Defining tool schemas and instructions so the AI knows when and how to retrieve CRM information. * Building secure retrieval mechanisms that enforce strict user and organization-level access controls. * Transforming raw CRM records into structured, AI-ready context. The AI will need to reason across: * CRM contacts and organizations * client profiles * Deals and opportunities * Projects * Tasks and reminders * Notes * Email history * SMS and WhatsApp communications * Call transcripts * Meeting summaries * Documents and contracts * Workflow history Agentic AI & Workflow Automation * Build proactive AI agents that generate alerts, recommendations, follow-ups, reports, and suggested next actions. * Design systems capable of reasoning across both structured and unstructured data sources. * Architect and implement multi-step and multi-agent workflows. * Develop workflow intelligence that assists users in completing real-world business tasks. Required Experience * Demonstrated experience building and deploying production AI systems used by real customers. * Experience working with embeddings, vector databases, and retrieval pipelines. * Experience implementing LLM tool/function-calling architectures. * Experience integrating AI systems with business systems such as CRMs, ERPs, or other operational databases. * Experience combining structured and unstructured data within AI applications. * Strong backend engineering and systems architecture experience. * Demonstrated ability to quickly understand and improve existing codebases. * Ability to independently own and deliver complex technical initiatives. Strongly Preferred * Experience with OpenAI embeddings. * Experience with MongoDB Atlas Vector Search. * Experience building agentic AI systems and workflow automation. * Experience designing long-term memory architectures. * Experience building multi-tenant SaaS applications with strict authorization requirements. * Experience implementing evaluation and monitoring pipelines for production AI systems. What We Value * High accountability and ownership. * Strong communication skills. * Product thinking and user empathy. * Ability to understand user workflows before writing code. * Pragmatism and sound engineering judgment. PLEASE DO NOT WASTE OUR TIME IF YOU NOT MEET THE REQUIREMENTS 

AI SYSTEMS ENGINEER Agentic AI, Multi-Agent Systems & Secure AI Workflows (U.S.) Remote • United States We're building production AI systems designed for enterprise environments. We're looking for exceptional AI systems engineers who enjoy solving difficult systems problems – not just writing code. Our work sits at the intersection of agentic AI, software architecture, enterprise systems, governance, security, and operational intelligence. We design AI systems that improve how organizations operate while meeting the standards required for production deployment. We value engineers who think in systems, challenge assumptions, and care deeply about building technology that is reliable, understandable, secure, and useful. If you're motivated by difficult engineering problems, thoughtful architecture, and building production AI systems for enterprise organizations, we'd like to hear from you. WHAT YOU'LL HELP BUILD Examples of the types of systems we design include: - Multi-agent AI systems - Enterprise AI assistants - Secure AI workflows - Enterprise workflow automation - AI-powered knowledge systems - Human-in-the-loop decision support - Document intelligence - Retrieval-Augmented Generation (RAG) - AI memory and retrieval systems - AI evaluation and testing frameworks - Secure enterprise AI platforms - AI governance capabilities - Operational intelligence platforms TECHNICAL EXPERIENCE WE VALUE We're interested in engineers with experience in some combination of: - Python - AI Agent Development - LangGraph - LangChain - Large Language Models - API Development - Vector Databases - Software Architecture - Enterprise Systems Integration - Information Security Experience with OpenAI, Anthropic, Model Context Protocol (MCP), cloud infrastructure, workflow orchestration, observability, distributed systems, or regulated technology environments is also valuable. We do not expect expertise in every technology. We care far more about engineering judgment, systems thinking, demonstrated execution, and continuous learning than checking every technology box. THE PROBLEMS WE ENJOY SOLVING The engineers who thrive here enjoy questions like: - How should multiple AI agents coordinate work? - How should humans remain in control of important decisions? - How should production AI systems scale safely? - How should memory be designed for enterprise AI? - How should AI systems balance operational performance with governance, security, and reliability? - How should AI systems create measurable business value? If those questions excite you, you'll probably enjoy working with us. WHAT MAKES SOMEONE SUCCESSFUL HERE We're looking for engineers who: - Think in systems rather than individual features. - Care deeply about production quality. - Enjoy solving ambiguous technical problems. - Communicate complex ideas clearly. - Balance speed with sound engineering judgment. - Build practical solutions rather than chasing hype. - Continuously learn, experiment, and improve. We're significantly more interested in systems you've built than technologies you've used. Please provide specific examples that demonstrate your role, engineering decisions, and measurable outcomes. We recognize that many engineers use AI as part of their workflow. You're welcome to do the same. However, your application should accurately reflect your own experience, judgment, and technical thinking. We respect the confidentiality of your current and former employers, clients, and partners. Please do not include proprietary or confidential information in your application. Describe your work at a level that demonstrates your engineering approach without disclosing protected information. PROFESSIONAL STANDARDS We value integrity, sound engineering judgment, and respect for intellectual property. Please do not include confidential, proprietary, export-controlled, or other non-public information belonging to your current or former employers, clients, or partners in your application or work samples. We're interested in your engineering approach, architectural thinking, and problem-solving methodology – not protected information belonging to others. If you share code, architecture diagrams, technical documentation, or project examples, please ensure you have the legal right to do so and identify any material open-source or third-party technologies where appropriate. By submitting application materials, you represent that you have the legal right to share them and that doing so does not violate any confidentiality, intellectual property, employment, consulting, or other contractual obligations. Any engagement, if offered, will be subject to a separate written agreement covering confidentiality, intellectual property ownership, compensation, and other applicable terms. Submission of an application or participation in the evaluation process does not create any employment, independent contractor, partnership, joint venture, agency, fiduciary, or other business relationship with 26ers AI, nor does it obligate either party to enter into any future engagement. 26ers AI reserves the right to evaluate applications, discontinue discussions, modify the hiring process, or decline to pursue any engagement at its discretion. Nothing in this posting should be construed as an offer of employment or an offer to contract.

1. Project Overview This project builds a computer vision system that tracks whether a handheld object is inside or outside a user defined region on a person's head in real time. The system observes the user through a webcam, allows the user to define a region on their head once at the start, tracks that region as the user moves, and outputs a binary signal whenever a tracked object is inside or outside the region. The signal is transmitted to an Arduino over USB serial and can be used to control any device, such as turning a flashlight on when the object is in the correct region and off when it leaves. The previous developer completed Milestone 1, which covers webcam capture, frame rate monitoring, multi camera support, and a basic overlay system. The remaining work is organized into five milestones over an estimated nine weeks. 2. System Architecture The system runs as a single Python application on the user's computer. It captures frames from the webcam, runs all computer vision processing, and sends commands to the Arduino over USB serial. The Arduino runs a minimal firmware that reads those commands and switches an output pin accordingly. That output pin can drive any device the client chooses. The processing pipeline has six stages that run on every captured frame. First, the frame is read from the camera and corrected for lens distortion using stored calibration parameters. Second, MediaPipe Face Mesh detects 468 facial landmarks on the user's head. Third, the previously stored region anchor points are reprojected into the current frame using the new landmark positions, which is how the region follows the head as it moves. Fourth, a color based tracker locates the handheld object in the frame. Fifth, the system computes whether the object is inside or outside the region using a point in polygon test. Sixth, the resulting on or off signal is sent over serial. A separate safety supervisor runs alongside the main pipeline. It forces the output signal to off if the face cannot be detected, if the object cannot be tracked, or if the serial connection is lost. This guarantees the output is only on when the system is confident about the object's position This WHAT Im looking to be Completed a simple part of the project: Milestone 1: Camera Setup & Calibration Goal: Set up the camera and calibrate the system for accurate real world distance measurements. Steps: Set up and test the camera for real time video capture. Correct camera distortion using OpenCV and a checkerboard pattern. Calibrate the system using a reference object of known size and adjust dynamically based on the user's inter pupillary distance. Success Criteria: Achieve real world measurements with an accuracy within 5%. Correct camera distortion and dynamically adjust scale during use. I do have existing code that does cover this part:with basic webcam capture, FPS display, multi camera preview and overlay logic It is however missing this:calibration, head anchored region tracking, smoothing, object detection, Kalman filtering, Arduino communication and safety logic are missing.

We are academic researchers studying how software engineering work is organized and how it has changed over time, particularly in response to generative AI tools. We are looking to interview several software engineers about their professional experience. Any information you provide will be used as part of academic research outputs, and will not be attributed to your name or your company. Compensation: $30 fixed rate (i.e., $60/hour for a 30-minute Zoom interview) --- Who we're looking for - Currently working or recently worked (within the past 6 months) as a full-time software engineer at a company. Since we are interested in learning about your usage of AI within a software engineering team, we cannot accept individuals who do not have this form of recent experience. - Any level of seniority welcome What the interview covers - How engineering work is organized at your company (or past company) - How workflows have changed in response to AI - Your experience with tools and workflows Output - The qualitative interviews will be used to inform research hypotheses for the academic project - With permission, we would also include the information directly in the qualitative section of an academic paper, covering interviews with 30+ tech workers --- e.g., as paraphrased or anonymized quotes --- To apply, please briefly describe: 1. Your experience in software engineering roles, over the past 4 years -- company names, job titles held, and dates of positions held 2. Your availability for a 30-minute video call interview Thank you!

Forum Intelligence: Project Brief & Initial Rollout 1. Executive Summary & Objective Forum Intelligence is a beginning as a localized data retrieval, processing, and archiving system designed to scrape public municipal records and state legislative data for public oversight. The immediate objective is to build a functional, highly resilient prototype focused on the Tri-Cities region (Burbank, Glendale, and Pasadena, California). The system will autonomously ingest messy, unstructured municipal data (City Council meeting minutes, agendas, public notices, and legislative PDF text, recorded mp4), clean it, and make it fully searchable and queryable via a localized AI agentic framework. 2. Phase 1 Scope: The Tri-Cities Rollout Th engineer will be responsible for building two primary pillars: A. Resilient Scraper Bots • Target Ingestion: Monitor and pull data from Burbank, Glendale, and Pasadena municipal portals and California legislative feeds. • Data Types: Brittle HTML sites, heavily nested tables, public notices, legislative drafts, and massive unstructured PDF archives. • Requirements: The scraping architecture must be exceptionally robust, utilizing intelligent error handling, retry semantics, and pagination tracking to handle frequent municipal website layout changes without breaking the pipeline. B. Ingestion & Vector Pipeline • Parsing: Extracting clean text from poorly formatted documents and scanned PDFs. • Local RAG (Retrieval-Augmented Generation): Chunking and embedding the data locally into a vector database (e.g., pgvector, Chroma, or Milvus) to enable semantically accurate entity linking and contextual search. 3. Targeted Hardware Stack To ensure maximum data security, strict public oversight integrity, and predictable operational costs, Forum Intelligence is skipping commercial cloud APIs in favor of an on-premise, localized NVIDIA enterprise deployment. The production roadmap aligns precisely with the new computing patterns detailed in NVIDIA’s latest hardware roadmap: • Inference & Token Generation: Running local open-weight frontier models (e.g., Neotron 3 Ultra or Claude/Llama equivalents) optimized for reasoning and long-context tool use. • Compute & Orchestration: The backend infrastructure is architected around NVIDIA’s dedicated agentic architecture, utilizing high-instructions-per-clock (IPC) Vera CPUs paired with Vera Rubin GPUs. • Memory & Storage Processing: Utilizing NVIDIA’s unified memory fabric and data processing units (DPUs) for ultra-low latency context management, KV caching, and fast vector database retrieval. 4. Immediate Milestones for the Engineer 1. Architecture Design: Map out the database schema and local inference ingestion loop. 2. Tri-Cities Scraper Deployment: Write and deploy the initial automated bots for Burbank, Glendale, and Pasadena. 3. Local MVP Pipeline: Demonstrate a local RAG pipeline where a user can query the Tri-Cities scraped records and receive grounded answers with exact source attributions. The above was AI generated from months long conversations with Gemini. The goal is to prove the concept then roll out to LA County, state of CA, and then the country.