🤓 Latest Submissions
Bob AutoHeal — AI Testing & Validation Hub
Problem Software development teams lose enormous time in a repetitive, manual debugging loop: write code, push, CI fails, read logs, guess the root cause, apply a fix, repeat. Each bug cycle typically takes 30–45 minutes of developer time — time spent reading stack traces, cross-referencing source code, and applying fixes one at a time. As codebases scale, the number of bugs outpaces the team's ability to resolve them quickly. This problem is especially acute in three common bug categories: missing guard clauses (e.g., unhandled KeyError causing HTTP 500), missing input validation (e.g., negative quantities silently accepted), and incorrect HTTP status codes used for business logic errors (e.g., returning 500 instead of 400 for out-of-stock). These bugs are subtle, pervasive, and expensive to find manually. Solution: Bob AutoHeal Bob AutoHeal is an Automated Testing and Validation Hub powered by IBM Bob 2.0. It transforms the manual debug loop into a fully autonomous, AI-orchestrated workflow. The system works as follows: 1. Validation: A developer triggers `python scripts/run_validation.py`. This script runs the full pytest suite, generates a code coverage XML report, and outputs a structured failure report — identifying exactly which tests fail and why. 2. Research: IBM Bob's Agent spawns a Research Subagent that reads the source code (`app.py`) and the pytest stack traces side-by-side using Document Understanding. It identifies the root cause for each failure with precision. 3. Parallel Healing: Bob spawns one Coder Subagent per bug, all running simultaneously using Parallel Tasks. Each subagent applies a targeted, minimal fix to the codebase without affecting unrelated code. This parallelism cuts total fix time by approximately 3x compared to sequential patching. 4. Verification: Bob re-runs the validation script. If all tests pass, the workflow completes and Bob generates an impact report documenting what was fixed, time saved, and coverage achieved. Target Users - Software developers and DevOps engineers who want to eliminate manual debugging from their CI/CD pipeline - Engineering teams maintaining large Python/FastAPI codebases - Hackathon participants demonstrating AI-assisted development practices What Makes It Unique Unlike simple code-completion tools, Bob AutoHeal operates as a full autonomous agent loop — it detects, diagnoses, fixes, and verifies, all without a developer needing to read a single stack trace. The parallel subagent architecture means multiple bugs are fixed simultaneously, not sequentially. The inclusion of a structured Bob Skill file (`bob_skills/AutoHealer_Skill.md`) makes the workflow reusable and teachable to any team. Impact In our demonstration, the AutoHeal agent resolved 3 bugs across 4 bug categories in under 2 minutes, achieving 8/8 tests passing with 100% code coverage. The equivalent manual process takes 45 minutes. This represents a 95% reduction in resolution time per bug-fix cycle.
27 Sep 2026
SentinelEdge: Bimanual VLA on Intel Core Ultra
SentinelEdge is a complete Physical AI system that tackles the three hardest bottlenecks in real-world robot deployment: multi-modal grounding, bimanual coordination, and edge inference latency. Two simulated SO-101 robotic arms operate inside a MuJoCo physics environment, executing complex multi-step dinner-table-setting tasks: opening drawers, retrieving objects, performing cross-arm hand-offs, and avoiding a dynamic sinusoidally oscillating obstacle. At the core is a Temporal Action Chunking Transformer (ACT) that fuses live camera pixels, real Sentence-Transformer language embeddings, and joint-action history through a 4-layer Cross-Modal Transformer, outputting independent 8-DOF trajectories for each arm. The policy is exported to Intel OpenVINO INT8 via NNCF Post-Training Quantization and benchmarked on Intel Core Ultra NPU at 85.34 FPS — 2.8× the headroom needed for 30Hz real-time control. Robustness is demonstrated across 10 randomized seeds with 6-axis domain randomization (placement, mass, friction, lighting, shape scale, floor texture). A Gradio dashboard with Speechmatics Batch ASR lets users speak commands that the robot executes and verbally confirms via Text-to-Speech. The project includes a multi-stage FSM reward system, 39-test pytest suite, ablation study notebook, HuggingFace Model Card, EMA kinematic action smoother for Sim2Real transfer, Foxglove Studio telemetry dashboard, and a full ROS 2 deployment node.
16 Sep 2026
AlphaLoop: Autonomous Neural Options Trader
The Problem Retail options traders are outgunned by institutional algorithmic desks. They lack the speed to parse complex signals (RSI, IV Rank) and the discipline to stick to rigid risk management rules. What it does AlphaLoop acts as a fully autonomous hedge fund manager. The system continuously scans 20 global markets. When an opportunity arises, it routes live data to a Gemini-powered Strategy Selector. Gemini analyzes market conditions and selects optimal multi-leg options strategies (e.g., Iron Condors for high-IV markets, Credit Spreads for momentum). Before execution, a rigid mathematical "Risk Guardian" agent intercepts the order to evaluate portfolio net delta, sizing constraints, and daily loss limits. If approved, the Execution Agent fires the trade directly to the Alpaca API via the Model Context Protocol (MCP). How we built it The backend is an asynchronous Python FastAPI server deployed on Render, connecting Gemini 2.5 Flash to Alpaca's trading infrastructure via MCP. Our frontend is a Bloomberg-inspired dashboard hosted on Vercel. To give judges transparency into the AI's "brain," we built real-time visualizers: a dynamic Chart.js Strategy Matrix, an Institutional Order Flow tracker, and a Web Speech API voice assistant that audibly announces the AI's decisions. Challenges & What's Next We overcame complex edge cases involving portfolio delta calculations triggering false positives in our risk gates. Next, we plan to add a RAG pipeline to parse SEC filings for earnings plays.
4 Sep 2026
