🚀 Solo founder + independent researcher
🧠 Building AI tools for developers and researchers
📄 1 research paper under review
I build small useful AI tools and open-source developer utilities.
- 📡 TraceLLM - Open-source LLM observability platform → track prompts, token usage, latency, retries, hallucinations, tool calls, agent execution paths. PostgreSQL stores traces. WebSocket streams logs live.
- ✅ CommitWrite — Reads your staged git diff and generates a clean commit message using AI
- 📜 PaperDigest — Paste any arXiv paper link → get structured summaries (problem, method, results) instantly
- 🐛 BugReport — Paste any error + stack trace → plain English diagnosis, exact fix, and prevention tip with severity badge
- 📊 TabExplain — CSV → visual dashboard (patterns, outliers, correlations)
- 🧹 DataClean — Upload messy CSV → auto-clean (missing values, types, duplicates) + visual before/after report
- ⚡BenchMark — Compare ML models on tabular datasets with live metrics.
- 🔬 TinyTrain — Fine-tune small ML models from CSV datasets.
- 🧪 AgentBench Lite — Benchmarking platform for evaluating LLM agents and tool usage.
• 🧠 VectorDoctor — Debug vector databases, retrieval quality, embedding collapse, and chunk overlap directly from terminal.
• 🛡️ PromptShield — Scan prompts for jailbreaks, prompt injection, hidden instructions, and unsafe tool access.
• ⚡ InferPulse — Benchmark Ollama, vLLM, TensorRT, ONNX, and llama.cpp latency, throughput, and VRAM usage.
• 🧬 SemanticDiff — AI-aware Git diff CLI that detects architectural impact, logic changes, and bug-risk.
• 🤖 AgentOps CLI — Trace multi-agent execution, retries, loops, tool calls, and memory failures in realtime.
• 📉 DriftWatch — Detect concept drift, feature shift, PSI divergence, and ML anomalies from streaming datasets.
• 🧩 ContextForge — Compress long-context conversations, optimize tokens, prune irrelevant memory, and summarize history.
• 🧪 EvalForge — Generate jailbreak prompts, adversarial datasets, multilingual evaluations, and reasoning benchmarks.
• 🖥️ GPUQueue — Queue AI jobs, manage GPUs, retry failures, and monitor utilization from terminal.
• 🔍 RepoGraph — Analyze huge repositories, dependency relationships, architecture flow, and code ownership.
• 📦 DockerPulse — Diagnose Docker performance, broken containers, memory leaks, and unhealthy services.
• 🌐 APIScope — Realtime API debugging CLI with latency tracing, error replay, and request inspection.
• 🔥 CacheDoctor — Redis and caching debugger for cache misses, stale data, hot keys, and memory spikes.
• 🧵 QueueWatch — Inspect Kafka, RabbitMQ, and Redis queues for stuck jobs, retry storms, and throughput bottlenecks.
• 🛠️ PromptCompiler — Convert natural-language tasks into structured prompt workflows with tools and retries.
• 🔐 SecretRadar — Scan repositories, containers, and environments for leaked API keys and secrets.
• 🛰️ ModelWatch — Monitor local/open-source models for latency spikes, VRAM overflow, and unstable inference.
• 📡 PacketLens — Network debugging CLI for API latency, packet drops, websocket failures, and DNS bottlenecks.
• 🧠 MemoryScope — AI agent memory debugger for episodic memory corruption, context overflow, and retrieval failures.
• 🧪 ChaosForge — Simulate GPU crashes, API failures, latency spikes, and broken model pipelines for resilience testing.
• Building open-source AI developer tools
• Publishing ML and AI research
• Experimenting with open-source models like Qwen
Data-Efficient Machine Learning for Small Tabular Datasets: A Comparison Study
- 💻 Code: https://github.com/avikcodes/data-efficient-ml-small-tabular
- 🔗 DOI: https://doi.org/10.5281/zenodo.19394292
Tool-Augmented AI Agents on Small toMedium Language Models
• I Trained 12 ML Models Across 4 Datasets With Zero GPU Cluster. The Results Destroyed My Assumptions
• Humanoid Robots & Physical AI are finally entering homes and blue collar jobs in 2026!
