Generative AI & LLMs
Bundesliga Wrapped
AWS World Sports Innovation Cup · Finalist (Germany)
A team project building "Bundesliga Wrapped" — a Spotify-Wrapped-style end-of-season recap of the league. My contribution: a Python pipeline that samples match-clip frames with OpenCV and calls Amazon Nova Lite via the AWS Bedrock Converse API to produce structured JSON on players, events, and goals that feeds the wrap's statistics.
Streamlit Clinician-Training Platform
EPFL LiGHT Lab
A clinician-training platform integrating OpenAI, Anthropic, and Qwen LLM APIs for case-based learning by doctors in Rwanda.
Deep Learning & Speech
Dynamic Perceptual Steering
CMU 11-785 · Team Project
Investigates whether language prompts can steer vision-language models (InternVL-3, CLIP) away from Western-centric interpretation toward African cultural recognition — measuring shape/texture bias and "perceptual erasure" across the SURA and Africa-500 datasets.
End-to-End Speech Recognition Transformer
CMU 11-785 · HW4P2 · Top 11%
Encoder–decoder Transformer (21.3M params, 6+6 layers, 8 heads, 5k BPE tokenizer) for end-to-end ASR — ranked 8/71 on the Kaggle competition.
Face Verification with ResNet34
CMU 11-785 · HW2P2
ResNet34 trained from scratch (25.7M params) for face verification across 8,631 identities, reaching 2.63% EER and beating the course reference benchmark.
Utterance-to-Phoneme Recognition with CTC
CMU 11-785 · HW3P2 · Top 24%
2-layer Conv1d → 3-layer BiLSTM (hidden 256) → 2× pBLSTM encoder with a CTC decoder for end-to-end speech-to-phoneme recognition. Ranked 60/257 on the Kaggle competition.
Phoneme Classification with MLP
CMU 11-785 · HW1P2 · Top 20%
8-layer fully-connected MLP with a diamond architecture (1024 → 2048 → … → 512 → 42 phonemes) for frame-level phoneme classification. Ranked 10/50 on the Kaggle competition.
NMF for Vocal Separation
MLSP HW3
NMF-based blind source separation for monaural audio — isolating vocals from accompaniment at ~2.82 dB SDR on MUSDB18-HQ. Includes sparse recovery via Iterative Hard Thresholding (IHT) and music-decomposition exercises.
Software Engineering
TaskFlow API
REST API · Clean Architecture
Portfolio REST API in .NET 8 / C# 12 with clean architecture (Repository, Unit of Work, DI). JWT auth with PBKDF2 hashing, role-based authorization, EF Core 8 code-first on SQL Server, Serilog, Swagger, xUnit/Moq tests, Docker, Azure-ready.
CareNest
Maternal Health App
React/TypeScript maternal-health application designed for the Rwandan healthcare context.
Finndex Africa
PropTech Platform
PropTech platform built with NestJS and MongoDB.
SafeSpeak
Mental Health App · Beta
Secure mental-health support and community app, currently in beta testing. Incubated under the Mastercard Foundation Scholars Entrepreneurship Fund.
Security & Data
Information Security Projects
CMU Privacy & Security
Secure coding, encryption, threat-modeling, and hands-on vulnerability assessments — including Intrusion Detection Systems and Snort lab work, from CMU's Usable Privacy & Security and IDS coursework.
Data Inference & Applied ML
CMU Coursework
ML implementations across classification, regression, clustering, and time-series tasks, from CMU's Data Inference & Applied ML course.
Research
Speech-Based Depression Detection
M.S. Thesis · CMU · Prof. Prasenjit Mitra & Prof. Rita Singh
Voice-based clinical depression detection reaching F1 0.917 (Yoruba) and 0.794 (English) with fine-tuned HuBERT and WavLM in PyTorch. Released YODEP — the first open bilingual English–Yoruba clinical speech corpus.
Clinical ML Pipeline for Hepatocellular Carcinoma
Summer Research Intern · EPFL LiGHT Lab
Survival ML (Cox, RSF, DeepSurv, N-MTLR) on a nine-centre, 2,630-patient HCC prognosis study, reaching OS C-index 0.752 and matching it (0.748) with a privacy-preserving federated Cox model. Sole engineer with CHUV surgeons; top 1.3% of Summer@EPFL applicants.
Publications
YODEP: A Bilingual English–Yoruba Clinical Speech Corpus for Depression Detection
The first open bilingual clinical speech corpus for depression detection. Peace Bakare et al.
A PRISMA 2020 Systematic Review of Speech-Based Depression Detection
Systematic review of 276 papers across 14 databases on speech biomarkers for depression. Peace Bakare et al.
-
Jul 2021 – Mar 2025
Backend Engineer
Sterling Bank / The Alternative Bank, Lagos
Architected the Signatory Portal — a Central Bank of Nigeria-mandated UBO & AML compliance platform — as sole backend engineer and project lead. Re-architected the Account Statement app (~4 min lower latency) and cut team-wide debug time 30% on SterlingPro (100,000+ active businesses).
-
Jan 2024 – Feb 2025
Software Engineer
Axxess Technologies, US (Remote)
Maintained Axxess HomeCare across 56 microservices on ASP.NET Core MVC for US healthcare clients — resolving 40+ feature and bug tickets via code changes and query optimisation traced through Kibana logs, across a five-country team.
-
Aug 2024 – May 2026
M.S. in Information Technology
Carnegie Mellon University · College of Engineering
Concentration in AI & Software Engineering and Privacy & Security. Coursework: Introduction to Deep Learning (Bhiksha Raj), Machine Learning for Signal Processing, Data Inference & Applied ML, Usable Privacy & Security.
-
2015 – 2019
B.S. in Computer Sciences
University of Lagos, Lagos
Articles coming soon
I'm putting together write-ups on generative AI, LLM pipelines, and speech ML. Check back shortly.
"Add a short quote from a manager, collaborator, or advisor here — a sentence or two about working with you."
"A second testimonial goes here. Keep each one concise so the cards stay balanced."
I'm an AI/ML software engineer based in Leeds, UK. I build generative-AI products — LLM pipelines on AWS Bedrock (Nova, Claude), OpenAI and Anthropic — and fine-tune speech models like HuBERT and WavLM, on top of five years of production backend engineering in C#/.NET, Java and Python inside a regulated bank. I hold an M.S. in Information Technology (AI & Software Engineering; Privacy & Security) from Carnegie Mellon University.
Download CV (PDF)Skills
Honors & Awards
AWS World Sports Innovation Cup — Finalist
Finalist (in Germany) for an AWS Bedrock-powered sports video-analytics build.
Upanzi Scholarship
Fully funded two-year graduate scholarship (≈ $120,000 tuition) covering the MSIT program at CMU Africa.
Scholars Entrepreneurship Fund 2.0 — Mastercard Foundation
Selected for incubation of SafeSpeak; up to $5,000 funding, six-month mentorship, and entrepreneurship training.
2nd Place — AIMS Knowledge Technology Workshop
Runner-up among competing teams at the African Institute for Mathematical Sciences workshop; awarded a monetary prize.
University of Lagos Scholar Award
Recognized for first-class academic performance with a merit scholarship honorarium.
Community Impact & Service
Volunteer Web Developer — LiGHT Lab, EPFL
Co-built a training website helping medical doctors in Rwanda learn machine learning concepts.
Technical Mentor — BuildGems
Mentored three undergraduates through hands-on software-engineering and ML projects and career development.
Curriculum Designer & Instructor — iLift Community
Designed an IT curriculum for Nigerian high-school graduates to build foundational computing skills.
Instructor — First Class Foundation, UNILAG
Taught Introduction to Computer Science using VB.NET to strengthen computing literacy among first-year students.
Get in touch
Open to AI/ML and software engineering roles, and happy to talk about generative-AI delivery, speech ML, or backend systems.
- pbakare@alumni.cmu.edu
- +44 7480 861985
- Leeds, United Kingdom
- UK High Potential Individual visa