I’m looking for an AI Engineer to help build an automated red-teaming product based on open-source models. This is a short-term, hands-on project for around 2 months, with an expected commitment of about 20 hours per week. The goal is to build a specialized red-teaming engine that can generate adversarial prompts across different risk domains, severity levels, and attack strategies — then automatically run those prompts against target AI models to identify bad cases, failure patterns, and safety gaps. 🔍 What you’ll work on Build red-teaming systems on top of open-source LLMs, including fine-tuning, prompt optimization, evaluation pipelines, and model orchestration. Design automated prompt generation workflows across risk domains such as self-harm, hate, violence, sexual safety, misinformation, fraud, cyber, and other high-risk areas. Generate prompts across different harm levels, from benign edge cases to policy-borderline and clearly unsafe scenarios, while maintaining structured taxonomies and evaluation criteria. Run automated tests against target models such as Gemma, Llama, Qwen, or other open-source / closed-source models to surface jailbreak patterns, over-refusal, under-refusal, and policy inconsistencies. Build feedback loops that turn model failures into stronger red-team prompts, improved eval sets, remediation recommendations, and continuous safety testing. 🧠 What I’m looking for Hands-on experience with open-source LLMs, fine-tuning, LoRA / QLoRA, RAG, model evaluation, and LLM inference pipelines. Familiarity with AI safety, red teaming, adversarial prompting, jailbreaks, safety evals, or trust & safety systems. Ability to build end-to-end systems, including data pipelines, model serving, eval harnesses, scoring, dashboards, and automation workflows. Bonus if you’ve worked on model safety, content moderation, policy evaluation, agentic testing, or automated eval infrastructure. ⏳ Project setup Duration: around 2 months Time commitment: about 20 hours per week Format: flexible / remote-friendly Stage: early-stage build, from 0 to 1 🚀 Why this is interesting This is not about manually writing red-team prompts one by one. The goal is to build a scalable system that can continuously generate, test, categorize, and learn from model failures — helping teams understand where AI models break, why they break, and how to improve them. If you enjoy working with open-source models, AI safety, red teaming, and fast 0-to-1 product building, I’d love to chat. Feel free to DM me if this sounds like you, or if you know someone who might be a good fit.

I want to build a private multi-model RAG-based Opportunity Intelligence Agent. It should support document ingestion, opportunity-specific workspaces, vector search, source citations, multi-model routing across OpenAI, Claude, Perplexity, and possibly DeepSeek, and generate strategic recommendations from both uploaded files and live web research. This is intended to become a reusable base agent capable of knowledge retrieval, web research, multi-model orchestration, document analysis, citation generation, and agent clonding and configuration. It will be used for analyzing & strategy development for project opportunities, responding to RFPs, and proposal assistance, as well as other applications.

We're building an internal AI system that runs entirely on our own hardware (no cloud inference) against our own company data. We have a working proof-of-concept and want to get the architecture right. We need an experienced consultant to review what we've built, pressure-test our decisions, and tell us where we're wrong. This is an advisory/validation role first — we have someone doing the hands-on work; what we want is a senior second opinion to make sure we're building this the right way. What we're running today: Inference: RTX 5090 (32GB, Blackwell), Ubuntu 24.04, running llama-server (llama.cpp + CUDA) serving Gemma 4 31B-it (Q4_K_M GGUF) at a 262,144 context window. Also hosts our MCP retrieval server, PostgreSQL, and Qdrant. Embeddings: separate machine with an RTX 3060 running vLLM serving Qwen3-Embedding-4B. RAG: hybrid retrieval — Postgres full-text search + Qdrant semantic search with RRF fusion, exposed through a custom MCP server with tool-calling. Data: ingesting our own internal operational data into Postgres + Qdrant. Planned stack: LiteLLM for model routing, n8n for automation, Open WebUI for the interface, Langfuse for observability, Vault or Infisical for secrets, Keycloak/Azure AD for SSO. What we need help with: Validating our two-machine split (inference vs. embeddings) and whether our VRAM/context budget holds up under real load — specifically whether a 256K context window is real and performant on a single 32GB card or just nominal. Model selection and routing strategy: which open-weight models for which tasks, and how to structure LiteLLM routes. RAG quality: chunking, embedding dimensionality, hybrid search tuning, reranking — making retrieval actually accurate on messy real-world data. Sanity-checking our overall architecture and telling us our blind spots. You should have done: Stood up local LLM inference in production — llama.cpp/llama-server and vLLM, not just Ollama on a laptop. You understand GGUF quantization (Q4_K_M, IQ-series), KV cache, KV-cache quantization, and how context length maps to actual VRAM consumption. Real fluency in GPU sizing math — given a model, a quant, and a context window, you can tell us whether it fits on a given card and what throughput to expect. Bonus if you've worked with Blackwell / sm_120a. Built production RAG — vector DBs (Qdrant, pgvector), hybrid search, RRF fusion, embedding model selection, reranking, evaluation. Worked with agentic/tool-calling systems and ideally MCP servers. Know the open-weight model landscape (Gemma, Qwen, Llama, Mistral, Phi, Nemotron, Hermes) and their licenses well enough to advise. Production ops: systemd, Docker, model gateways (LiteLLM or similar), observability (Langfuse), secrets management, SSO.

We are seeking a US-based computer vision and full stack developer to build a platform for sports card recognition. The project includes developing subscription management, dashboards, and user account features. The ideal candidate will have experience creating scalable applications and integrating computer vision capabilities into a user-friendly platform. Hiring: Computer Vision + Full Stack Developer for Sports Card Live Auction Overlay App (SaaS) 📌 Overview I’ve built an MVP of a real-time sports trading card scanning and comping overlay tool using Loveable.dev. The product helps buyers gain an edge during live auctions by instantly identifying cards and showing real-time market comps. Now I’m looking for a U.S.-based developer (or strong US-aligned freelancer) to take this from MVP → production SaaS. This is a subscription-based product, so I need someone who can help build something fast, accurate, scalable, and hard to replicate. 🧠 What the product does Users can: Capture or upload sports trading card images during live auctions (mobile + desktop) Instantly identify: Player Year / set Parallel / serial number Pull live market comps Display a real-time “buy / avoid / fair price” overlay The goal is speed + accuracy in live buying situations (seconds matter). ⚙️ What I already have MVP built in Loveable.dev Basic overlay + UI flow Initial comp logic concept Subscription idea (not yet fully implemented) 🛠️ What I need help building (Phase 1 → Scale) I’m looking for someone to help rebuild and harden the system into a real SaaS product: 1. Computer Vision / OCR Layer Card detection from images (mobile + desktop) OCR extraction (player name, set, serial numbers) Image recognition / matching to known cards Confidence scoring (very important — must avoid wrong matches) 2. Comp Engine (Core Value) Integrate or build system for: eBay sold listings 130point or similar comp sources Card Ladder / ALT-style pricing logic Return: last sale average comp trend direction liquidity estimate 3. Real-Time Overlay System Lightweight overlay that works during live auctions Low latency (fast lookup is critical) Works on mobile + desktop workflows 4. SaaS Infrastructure User accounts + authentication Subscription billing (Stripe) Usage tracking / rate limiting Admin dashboard 5. Scaling / Production Hardening API architecture improvements Database structure Performance optimization for real-time use Error handling for imperfect images 💡 Ideal candidate You should have experience with: Computer vision (OpenCV, YOLO, or similar) OCR pipelines AI image classification or similarity matching Full-stack SaaS development Stripe subscriptions API design (Node.js / Python / Next.js preferred) Huge plus if you have: Sports card / collectibles knowledge Experience with marketplaces or scraping pricing data Real-time / low-latency systems 🎯 Why this is interesting This is not a generic app. It’s: A real-time decision engine for high-value collectibles Built for a passionate, high-spend niche (sports cards) Subscription-based with strong monetization potential Designed for speed advantage in live auctions 📍 Requirements Must be U.S.-based (preferred for communication/time zone alignment) Must be able to work independently Must have strong GitHub/code examples Bonus if you’ve built AI or vision-based SaaS tools before 💰 Budget Open to: Hourly or fixed project 📩 To apply, please include: Relevant CV / GitHub Past AI / computer vision projects Any SaaS or startup experience Your approach to building a real-time image → comp system Availability per week

We are seeking a senior AI developer to build and enhance AI models for our business. The role involves developing, testing, and deploying AI solutions, as well as improving existing models to increase accuracy and performance. The ideal candidate should have strong experience in AI development and be able to work independently on complex projects.

We need a senior architect to design and build a multi-model routing control plane, then lead a small senior team through the build. The control plane sits in front of a family of AI systems and decides, per request (text, image, video), the optimal path across cost, quality, latency, business value, and sovereignty (data residency, rights, and cultural fit): cache and reuse, a small or on-device model, an open-weight, fine-tuned, or sovereign model, or a higher-cost frontier fallback. It routes across compute too: CPU, GPU, inference accelerators, on-device, and edge. Core KPIs: the share of eligible workload kept off frontier accelerators and the resulting cost reduction on a representative workload, plus sovereignty compliance, with no quality regression. This is not a chatbot and not a wrapper over hosted APIs. You own the architecture, define the routing logic, and lead execution. You think in systems, not individual model calls. Context The router is one component of a larger AI platform. It must be model-agnostic: open-weight, fine-tuned, and proprietary models swap in and out behind a stable interface without rearchitecting. A separate team owns the models you route to. The engagement is a 60 to 90 day POC with a working router demo (text-first, with a defined path to image and video), followed by technical leadership through the build. What you'll own Control plane: intake and normalization, classification, routing taxonomy, model-selection logic, fallback hierarchy, cache and reuse rules, telemetry, and the eval feedback loop. Routing that is learned and calibrated, not just static rules: predict per-query difficulty and expected quality, and escalate on confidence thresholds. Comfort with cascades and speculative decoding is expected. Routing across cost, quality, latency, and policy. In constrained environments some requests must stay local regardless of cost. Model-agnostic interface: clean, stable contracts so models and execution paths swap without rework, and the separate model team can work independently of the routing layer. Cost optimization across compute: exact and semantic cache, prefix/KV cache reuse, output reuse, batching, small-model routing, CPU offload, and on-device/edge execution, with a clear fallback hierarchy. The goal is to move most eligible workload off frontier accelerators without degrading output. Generative caching and reuse: caching text is easy; image and video are not, since the same prompt should produce variation rather than an identical result. We need credible reuse at the asset or component level, not just for text. Eval loop: scores output quality by domain and flags weakness so the training team can target fixes instead of retraining broadly. Track quality vs intent, failure modes, cost per route, latency per route, cache hit rate, fallback rate, and regeneration rate. Execution and leadership: architecture blueprint, POC scope, milestones, infra assumptions, and risks leadership can review, plus hands-on architecture review and task breakdown. You'll lead a small senior team, and one of your first deliverables is recommending its exact composition (see screening questions). Ideal background Led or architected production AI infrastructure across several of: multi-model orchestration and LLM routing, multimodal, model serving, inference cost and GPU reduction, CPU and on-device inference, open-source and fine-tuned deployment, cascades and speculative decoding, semantic and prefix caching, eval pipelines, and AI observability. Deployed in at least one constrained environment: on-prem, self-hosted, air-gapped, or data-residency-restricted. You know what breaks when you can't lean on a single cloud. Can lead: set architecture, break down work, review the team's output, and keep the build on track. Tools matter less than the ability to architect the system correctly and lead execution. Not a fit: basic chatbot workflows, hosted APIs only, or prompt engineering alone. Deliverables Control plane blueprint, routing taxonomy, POC plan with milestones and success criteria, and an eval/feedback framework, with a working router demo as the 60 to 90 day target, then technical leadership of a small team through the build. Screening questions The most relevant AI routing, model-serving, or inference infrastructure system you personally designed or built: what was routed, which models or execution paths, and what role did you own? How would you design a router that chooses between cache/reuse, a smaller or local model, an open-weight or fine-tuned model, or a frontier fallback, across CPU and GPU? Where do learned routing, cascades, or speculative decoding fit? For generative image or video requests, how would you approach caching or reuse when the same prompt should still allow variation? Be specific. What metrics and eval loop would you use to prove the router cuts cost without degrading quality, and to help a separate training team find weaknesses? Beyond yourself, what team would you staff to hit these deliverables in 8 weeks? Give the roles, seniority, and headcount, how you'd split the work, and flag any deliverable that 8 weeks and a team of roughly 4 engineers can't realistically cover. To apply Answer the five questions, summarize your most relevant routing or inference-infrastructure work (repos, writeups, talks, or architecture you can describe), and give your high-level approach to a control plane that routes across cost, quality, and sovereignty while preserving quality. Note your availability, your rate, whether you've led a small engineering team before, and the team you'd staff to hit the deliverables in 8 weeks.

We are hiring an AI Engineer with strong hands-on experience building and shipping real AI products. Requirement: If you don't have a GitHub profile to share, this role is not a fit. What we’re looking for: • Strong experience in AI/ML engineering • Ability to build, test, and deploy production-ready AI systems • Practical experience working on real-world AI projects To apply: Please share your portfolio, past AI projects, and relevant work samples. Applicants without portfolio will be ignored.

We’re hiring an extraordinary developer to own and grow our Base44 apps and sales products. around the future of AI discovery 1. Future of AI Discovery Core Demo – https://pull-discovery-core.base44.app/ You’ll evolve https://pull-discovery-core.base44.app/ into a beautiful, fluid, high‑performance, full-functional future of AI discovery demo following our advanced and sophisticated technical blueprint Integrate and orchestrate AI models incorporating LLM's, Search and World Models into a seamless experience with no visible seams between UX and intelligence. Own front‑end performance, responsiveness, and micro‑interactions—animations, transitions, and state changes should feel intentional and “alive,” not bolted on. Implement robust logging and analytics to understand how users explore, where they get stuck, and how the discovery engine can adapt dynamically. 2. Book Sales Engine – Six‑Channel Publishing System The second current Base44 project is a system that operationalizes our comprehensive sales plan across six channels. SEE THE COMPREHENSIVE BOOKSALES PLAN ATTACHMENT UNDERNEATH THIS POSTING You will: Translate a detailed multi‑channel publishing strategy (KDP optimization, physical bookstores via IngramSpark, other digital platforms, libraries, bulk institutional sales, and authority‑engine content marketing) into concrete workflows, tools, and dashboards. Build internal interfaces and automations to: Track metadata, pricing, and promotions across Amazon KDP and other platforms. Monitor campaigns across TikTok, Meta, LinkedIn, YouTube, newsletters, and partnerships. Surface KPIs like BSR, review velocity, ad spend, email growth, library adoptions, and bulk orders in a single, coherent view. Design light internal UIs that make it easy for non‑technical team members to update copy, add titles, trigger campaigns, and view performance without breaking anything. Implement robust, testable integrations between Base44, external APIs, and data sources to keep everything in sync as we scale from 8 to 22+ titles and beyond. Who You Are We’re not looking for a generic “full‑stack dev.” We’re looking for an unusual combination of visionary and doer: Creative technologist mindset – You think in systems and interfaces at the same time. You care deeply about how a product feels as well as how it works. Obsessed with execution – You’re disciplined, structured, and relentless about shipping. You break ambiguity into sprints, reduce complexity into tickets, and never let projects stall. Proactive owner – You don’t wait for instructions. You propose better ways to do things, flag risks early, and bring options—not problems—to every conversation. Strong product sense – You can balance ideal UX with realistic constraints and understand when to ship v1 vs. when to invest in polish. Comfortable with complexity – Multi‑channel distribution, layered data flows, and evolving requirements don’t scare you; they energize you. Ideal Skills & Experience You don’t need all of these, but you should recognize yourself in most: 5+ years building production web applications, ideally with a strong front‑end/UI focus. Deep experience with modern web stacks (React/Vue/Svelte or similar) and TypeScript, plus comfort with Node or comparable back‑end runtimes. Strong visual/UI instincts: experience collaborating with designers or owning design yourself for data‑rich interfaces and dashboards. Experience integrating AI/LLM APIs and retrieval systems into real products (RAG flows, multi‑step tool use, chat‑like interfaces, recommendation engines). Experience with analytics and experimentation: event tracking, funnel analysis, A/B testing. Familiarity with publishing, ecommerce, or multi‑channel marketing systems is a plus (KDP, IngramSpark, email platforms, ad platforms, analytics). Prior work in environments like Base44 or other low‑code/agentic platforms is a strong plus, but not required if you learn fast.

I am seeking an experienced ML engineer to provide insights on the design of a model I am planning to build. Your expertise in model design and architecture will be invaluable in helping me make informed decisions.

I'm an independent inventor (Massachusetts LLC, patent-pending) developing a portable sports-training device that uses a projected laser line and a global-shutter mono camera to measure the angle of a small metal striking surface at the moment of impact. Target accuracy is ±0.5° on the angular measurement, with measurement latency under 2 seconds, and direct-sunlight robustness as a key engineering risk. This is an end-to-end Phase-0 feasibility engagement. The working assumption is laser-line + global-shutter mono camera with bandpass filtering, but I want your read on whether that's the right approach for this accuracy and these conditions. I have a strong lean, not a closed decision, and I'd rather you push back early than build something the wrong way. Once we align on the approach, you'll spec the bench rig (camera model, laser modules, filters, optics, baseline geometry, target mounting); I'll source the parts from your BOM and either ship the components for you to assemble or assemble and ship a built rig, your preference, whichever fits your workflow best. From there you capture data under controlled and outdoor conditions, develop the detection and calibration pipeline, and deliver a working codebase plus a written accuracy/robustness report. Hardware is returned to me on completion (or retained for a follow-on engagement if we both want to continue). What you'll deliver: 0. Approach review + rig spec. A short written deliverable (2–4 pages) covering: (a) your read on the proposed sensing approach, affirm + refine, or argue for an alternative with reasoning and a specific recommendation; (b) a bench-rig BOM with specific parts (camera model, laser modules, bandpass filters, optics, mounting, target plate) sized for the working distance and accuracy spec; (c) laser-to-camera baseline geometry with your reasoning, and recommended calibration targets. I'll source the parts from your BOM. We'll decide together whether I ship components for you to assemble or assemble and ship a built rig, whichever you'd rather. 1. Rig assembly or acceptance + baseline capture. Receive shipped parts (or built rig), assemble or validate alignment as appropriate, confirm basic optical performance against the M0 spec, then capture a baseline dataset (~200 frames per configuration) under controlled indoor lighting. Photos of the as-built rig and a setup diagram included. 2. Detection pipeline. A Python/OpenCV module that extracts the projected laser line with sub-pixel accuracy from frames at 60–100 fps. Sub-pixel line fit (Steger, Gaussian, parabolic) or weighted centroid, your choice with a short justification. 3. Calibration framework. Documented procedure and accompanying script for mapping pixel displacement to angular displacement of the target plate, accounting for camera intrinsics, lens distortion, and laser-to-camera baseline geometry. Validation against ground-truth rig angles. 4. Robustness data capture + analysis. Re-capture under (a) bright indoor with mixed daylight and (b) direct outdoor sunlight, for both laser variants with and without matched bandpass filters. Quantified accuracy + jitter per condition. 4–8 page PDF report comparing visible-red + bandpass vs. near-IR + matched bandpass. 5. Stretch (optional milestone): First cut at deriving angle-at-impact from a short pre/post-impact image sequence, pseudocode or working prototype, whichever fits the time budget. Deliverable format: Well-commented Python module(s) in a Git repo I'll provide, a README that walks a junior engineer through running the pipeline end-to-end, the captured datasets (raw frames + ground-truth angles), and a PDF report. What I'm looking for: - Comfort giving an unambiguous engineering recommendation: "use this approach with these parts" or "don't and here's why, and here's what to do instead." Phase 0 succeeds or fails based on the judgment in Milestone 0 as much as the algorithm in later milestones. - 5+ years of practical computer vision work, with shipped projects involving line/edge detection, sub-pixel feature localization, or structured-light triangulation. - Comfort doing your own benchtop work; mounting, alignment, basic optics handling. - Strong Python + OpenCV; comfort with NumPy/SciPy for the line-fit and calibration math. - Camera calibration experience (OpenCV calibrateCamera, distortion coefficients, projective geometry). - A workspace where you can run an outdoor sunlight test safely and legally with a Class-2 visible-red laser and a Class-1 IR laser module. - Bonus: prior work with laser triangulation, structured-light scanning, or sports/motion-tracking applications. - Bonus: experience deploying CV pipelines to Raspberry Pi or ESP32-S3-class hardware (potential follow-on scope). Engagement: - Fixed-price (preferred): $5,000–$7,500 total, paid across 5 milestones (approach review + rig spec → baseline capture → detection pipeline → calibration → robustness report). - Hourly alternative: $70–$140/hr with a 75-hour cap, then re-scope. - Duration: 5–7 calendar weeks (approach-review phase happens up front; ~1 week round-trip shipping after rig build). - Weekly 30-min check-ins (US Eastern preferred; flexible). - Hardware: shipped to you fully insured at my cost. Returned (insured, my prepaid label) on completion, or retained for follow-on engagement. - Possible follow-on: porting the pipeline to Raspberry Pi / ESP32-S3, IR laser variant tuning, integration support for the next prototype phase. Before we start: Short NDA + IP assignment signed before I ship the kit, share the technical design doc, or grant repo access. Upwork's standard terms transfer IP on payment, but I want a standalone signed PIIA on file as well, routine, less than 1 hour of your time. To apply, please include: 1. 1–2 examples of prior CV work involving sub-pixel localization, line fitting, or laser/structured-light triangulation. Paragraph + GitHub or paper link. 2. Three or four sentences on your approach to extracting a sub-pixel laser line centroid from a single frame. 3. Confirm you have a workspace where you can run both indoor and outdoor (direct-sunlight) image captures with a small bench rig, and that you're comfortable assembling the shipped kit. 4. Whether you prefer fixed-price or hourly, and your proposed milestone breakdown. 5. Without committing to a final answer until you've seen the full spec, a quick take: do you think projected laser line + global-shutter mono camera is the right sensing approach for ±0.5° angular accuracy at 60–100 fps under direct sunlight, or would you steer me toward a different approach? Two or three sentences. Looking forward to talking with strong candidates. Jason