
Every month, Nigerian households juggle rent, DSTV, MTN data and generator fuel against a small naira cash buffer, while family abroad sends dollars and the USD/NGN rate keeps moving. People either check the rate all week or miss a shortfall until the fuel runs out. What's missing is a simple signal: do I need to act today, or not? BobNairaAssist is a small, explainable money assistant that answers that question. One pure Python function, `evaluate_buffer`, runs a priority waterfall and returns one of five outcomes, each with a plain-English reason. It stays **quiet** when the buffer covers the bills and FX is calm. It sends **ping_shortfall** when the buffer can't cover bills due within 30 days, naming the urgent ones due within 3 days. It sends **ping_fx_watch** when the naira weakens and nothing is planned. For a planned remittance it judges the rate from the sender's side: after a sharp naira weakening each dollar buys more naira, so it gives **suggest_send_now** ("the move may revert"); after a sharp strengthening it gives **suggest_send_later**; with a stable rate, **suggest_send_now**. The hints are advisory and say so. The judge path runs fully offline in DEMO_MODE with mock bills and FX, and needs no API keys. It is available as a Streamlit app (Judge demo, Playground and About Bob tabs), as a CLI, and as a live in-browser build of the same `app.py` via stlite. The repo has 42 pytest tests. IBM Bob IDE is the core of how this was built. The first offline scaffold was written before I had Bob access, and I don't claim it as Bob output. After that, I ran four real Bob tasks, and their exports are in `bob_sessions/`: - **task-01, read-only review** (0.216 Bobcoins): Bob read the whole repo, explained the architecture and found three decision-logic gaps. The most serious: "send now" fired on any negative rate move, including a sharp naira rally. - **task-02, fix** (1.01 Bobcoins): Bob added the new `suggest_send_later` outcome, bill-horizon and urgent-bill logic, and made a $0 remittance count as none. It wrote 17 new tests (19 -> 36 passing). - **task-03, ship** (1.89 Bobcoins): Bob wired the new outcome into the Streamlit UI, added a scenario helper plus 4 tests (40 passing) and refreshed the docs. - **task-04, correct** (2.20 Bobcoins): reviewing the demo from the sender's side showed beat 3 was backwards: it said "wait" after the naira weakened 5%, when each dollar buys more naira. Bob fixed the logic so that case now suggests sending now (or splitting), deprecated `suggest_wait`, relabelled the beat and updated tests and docs (42 passing). That is review, fix, ship, correct in about 5.3 of 40 Bobcoins. Next steps: a live FX adapter behind a flag, optional watsonx.ai phrasing in English or Pidgin, and WhatsApp/SMS pings. This is a proof of concept on mock data; I don't claim any user metrics.
27 Sep 2026

VoiceRemitNG helps people in Nigeria get practical remittance answers by voice. AssemblyAI Realtime Speech-to-Text, together with upload STT for recorded audio, turns spoken questions into text. A small agent routes requests to focused tools for fee comparison, foreign-exchange checks, send-now advice, and a receive-money checklist. The conversation is designed for clear, actionable guidance rather than a generic chatbot experience. DEMO_MODE runs without API keys so judges can explore the core flow reliably; live mode uses ASSEMBLYAI_API_KEY for AssemblyAI services. The project is built in Python with Streamlit and demonstrates an accessible voice-first experience for remittance decisions. Built for the AssemblyAI Voice Agent Hackathon on lablab.ai.
30 Sep 2026

EdgeRemit helps people in Nigeria decide when and how to send or receive remittances. The Streamlit and CLI agent combines fee and FX tools with an Intel OpenVINO CPU intent router that classifies user asks (rate check, fee compare, send-now advice, checklist) at the edge before tool execution. DEMO_MODE runs without API keys; LIVE_OPENVINO uses real OpenVINO Runtime on CPU. Built for the AI Infra Summit Intel Online track to show practical edge AI infra for everyday money decisions. The repository includes a reproducible local demo, clear setup instructions, and a simple workflow designed for reliable, privacy-conscious decisions on everyday transfers.
16 Sep 2026