Skip to content
View OsasDTEch's full-sized avatar

Block or report OsasDTEch

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
OsasDTEch/README.md

Wisdom Omons

AI engineer in Lagos. I build production voice agents, retrieval systems and multi agent workflows, and I measure before I tune.

Don't read my CV. Interview my agent: askwisdom.duckdns.org is a voice AI that knows my work, answers with facts only, and books a call with the human me. I built it, which is sort of the point.

Live systems, click through

System What it does
HeatDesk Voice agent answering out of hours calls for service trades. Qualifies the job, checks the service area, books the appointment. Emergency keywords escalate in code before the LLM ever responds.
DispatchIQ Multi agent voice system for call routing and escalation. LiveKit native agent transfers.
askwisdom The agent version of me. Grounded retrieval over one knowledge file, deterministic tripwires, Ollama with Groq failover.
PostWeek Learns a writer's voice from their posts, drafts a week of LinkedIn and X content from live trend research.
Style Search Meaning based hairstyle search. Hybrid dense plus BM25 retrieval with a facet layer that actually knows hair. Built in one weekend for a startup I admired, before they asked.
Logsu Digital SIWES logbook tracker for students in Nigeria.
voice-latency-profiler Open source. Per stage timing for voice agents: endpointing, STT, LLM, TTS, with p50 and p95. I assumed my LLM was the bottleneck. It was third slowest at 0.77s while STT finalization sat at 1.90s median. Measure first.

How I work

Evaluation harness before the pipeline. Deterministic code before probabilistic models. One knowledge file as the source of truth. My favorite bug story is the one where my own score dropped from 96 to 70 on unseen data, because fixing that overfit properly (regex tiers before the model, back to 84 with all safety gates passing) taught me more than the 96 did.

Stack: Python, FastAPI, LiveKit, Deepgram, LangGraph, PydanticAI, Qdrant, Neo4j, PostgreSQL, React, Docker, AWS.

Reach me

Open to full time or contract with small teams building something real.

askwisdom.duckdns.org · LinkedIn · book a call · omonswisdom.ict@gmail.com

Pinned Loading

  1. dispatchiq dispatchiq Public

    Python

  2. postweek postweek Public

    TypeScript

  3. voice-latency-profiler voice-latency-profiler Public

    Python