Loading...
Loading...
Internal tools and client builds that demonstrate the technical depth and end-to-end delivery capability behind every engagement. Ranked by what most accurately shows what we ship today.
Consulting-grade company analysis in 5–8 minutes — multi-agent due diligence with branded PDF output and a GO/NO-GO verdict.
Claude Code skill that dispatches 5–6 parallel subagents — web scraper, Perplexity researcher, financial data, regulatory (SEC/EDGAR), sentiment, and a crypto-only dev-activity agent — across 15 scoring modules with a traffic-light criteria and weighted risk tiers. Produces a consulting-grade branded PDF and an Obsidian vault note with a GO/NO-GO verdict. Handles both equity and crypto targets with type-aware criteria for L1 / L2 / DeFi / Memecoin / AI-DePIN / Infrastructure tokens. Cross-module synthesis surfaces compounding risks; valuation pass adds price-target scenarios; macro overlay covers 10 geopolitical and economic dimensions.
Daily multi-source intelligence pipeline replacing 30–60 minutes of manual portfolio monitoring.
16-step pipeline that fetches prices, on-chain metrics (TVL, fees, revenue via DefiLlama), macro data (FRED: fed funds, CPI, DXY, yield curve), and developer activity (Electric Capital + GitHub) for 70+ crypto assets. Rates each asset 1–10 using a token-type-aware prompt (L1 / L2 / DeFi / Memecoin / AI-DePIN / treasury), stores ratings and snapshots in SQLite for trend analysis, and emails a Bloomberg-style HTML briefing at 07:00 SAST. Friday weekly portfolio health report layers in sector exposure, concentration, and correlation analysis.
Production-grade RAG pipeline that ingests audit programs, runs static analysis, and surfaces findings against a 2,000+ historical-finding corpus.
URL-driven pipeline that ingests Immunefi / Code4rena / Sherlock / Cantina audit programs, runs a multi-tool static analysis stack (Slither, Aderyn, Wake), and cross-references findings against a 2,000+ Solodit corpus via Qdrant vector search. AI triage layer applies severity scoring with deduplication before findings reach a human reviewer. Methodology gate: every PR is categorised, recall@15 is measured against a held-out partition, and no release ships without meeting Δ thresholds.
Sponsor-prospecting pipeline for a crypto media BD team — scrapes creator channels hourly, extracts deals via LLM, ranks leads.
Python worker polls YouTube channel feeds hourly, diffs against a videos table, and runs LLM extraction on video descriptions to pull structured sponsor data — name, domain, affiliate URL, promo code, blurb, confidence. Sponsors are matched against an existing roster and classified as new or known. A Next.js dashboard and a 07:00 CAT email surface new prospects, ranked by spend signal: more channels carrying a sponsor = bigger budget = higher-priority lead.
A live product anyone can use — built, deployed, and generating leads.
Full-stack audit platform with a free 7-domain tier (lead generation) and paid 10-domain tier with keyword rankings, competitor analysis, and a prioritised action plan. Email confirmation flow, PayFast payment integration, branded PDF report generation, and geo-aware pricing (ZAR vs USD).
AI agents with persistent memory, proactive coaching cycles, and cross-bot intelligence.
Multi-agent AI platform with 4 specialised bots sharing a semantic knowledge base (340+ documents). Four-layer memory architecture (working, long-term, patterns, plans). Weekly intelligence cycle: Sunday cross-bot review with prep briefs, knowledge gaps, and guided questions. Wednesday commitment check-ins. Automated post-session extraction. Two interaction modes: Telegram and LiveKit voice calls.
Built for a client who needed one view across all their exchange accounts.
Live dashboard pulling open positions, wallet balances, and account equity across multiple accounts on Bybit, Binance, and Gate.io. Data writes to a shared Google Sheet with separate tabs for wallet balances, active positions, account equity, and a summary view. Includes CryptoPunks NFT floor price tracking via OpenSea with multi-tier API fallback. Runs on a 15-minute schedule on a dedicated VPS.
| 01 | 02 | 03 | 04 | 05 | 06 | 07 | |
|---|---|---|---|---|---|---|---|
| Scale | 15 scoring modules, 5–6 parallel subagents | 70+ assets, 16-step nightly pipeline | 2,000+ findings, 4 audit platforms | 40–50 channels, hourly ingest | 10 audit domains, 200+ sites processed | 4-agent system, 5 containers | 3 exchanges, multiple accounts each |
| Complexity | Type-aware criteria, cross-module synthesis, PDF generation | Multi-source signal aggregation, LLM rating, weekly report | Static analysis, RAG retrieval, AI triage | RSS diffing, LLM extraction, ranked dashboard | Payment flow, PDF generation, geo-pricing | Voice AI, multi-agent coordination | Multi-exchange APIs, NFT price tracking |
| Infrastructure | Claude Code skill, ReportLab PDF, edgartools, multiple data APIs | DigitalOcean droplet, daily cron, SQLite store | Local Qdrant, Ollama, Foundry, Etherscan | Docker worker + Vercel dashboard, Supabase | Next.js + FastAPI, Vercel + VPS | Docker Compose, LiveKit, Telegram | VPS, Google Sheets, 15-min schedule |
| Key AI capability | Parallel multi-agent orchestration with type-aware scoring | Token-type-aware rating + signal weighting | RAG retrieval against historical findings | Structured LLM extraction at scale | AI-generated audit reports at scale | Real-time voice + cross-bot context | Multi-tier API fallback + data aggregation |
| Demonstrates | Agent orchestration + consulting-grade output | Quantitative discipline + signal design | Production RAG + evaluation rigour | Full-stack scraping + enrichment + UI | End-to-end product delivery | Custom agent architecture | Client delivery + exchange integration |
Every engagement starts with a free 30-minute discovery call. We map your processes, identify the highest-ROI automation, and tell you exactly what it would take to build.
Book a free audit