>_ AB

Available for high-impact engineering problems

Abraham Babalola

Engineer

I build automated systems, security tooling, and data pipelines that remove manual work and catch risk early. Grounded in shipping reliable software that ties directly to a business problem, not just a technical one.

Explore Work ↓ 📄 Resume (PDF)

Featured Work

Systems & Research

Dynamic project cards with interactive architecture and engineering trade-offs.

Quantitative Systems & Risk

EdgeRunner

</> Source

Automated trading pipeline that compares live market odds to historical edge distributions, discards low-conviction trades, and halts at hardcoded drawdown limits, removing emotional decisions. Proven in live execution; dry-run mode validates logic with zero exposure.

Automated pipeline comparing live market odds against historical edge distributions to discard low-conviction plays before execution.

Execution Modes: Offline mock settlement, dry-run validation, and live adapter

Market Discovery

→

Edge Matcher

→

Risk Controls

→

Settlement

Synchronizes cross-league timing and combines candidate tickets across multiple rounds into a master board.

Circuit Breakers: Hard exposure constraints and automated abort triggers halt candidate execution under defined adverse loss conditions to protect capital.

Application Security & Fuzzing

Input Mutator

</> Source

Burp plugin, desktop GUI, and CLI that generates parser-aware payload mutations, exposing WAF bypasses and proxy/backend desyncs in minutes instead of hours of manual testing.

Instead of blind byte fuzzing that gets rejected at the network edge, it understands grammar and encoding layers to test whitelist boundaries and blacklist evasion.

Testing Intents: Whitelist probing, blacklist evasion, and proxy differential gaps

Target Input

→

Character Mutation

→

Representation

→

Transport Encoding

3-stage composition pipeline exploiting Unicode normalization, overlong UTF-8, radix variation, and delimiter ambiguity.

Deployment: Single fat JAR running as a native Burp Suite tab, a desktop GUI with a live normalization inspector, and a headless CLI pipeable to wordlists.

Market Intelligence & Consensus

Forecast Hub

</> Source

Clusters live Polymarket and Kalshi contracts into unified event-level consensus dashboard to aid with making business decisions. Backed by SQLite (WAL) for fast concurrent reads.

Filters out meme-coin speculative noise, normalizes disparate exchange schemas, and aligns identical targets side-by-side to flag real operational risk when market consensus splits.

Persistence: SQLite in WAL mode (sub-millisecond reads, zero-lock concurrency)

HTTP/2 Streams

→

Noise Filter

→

Two-Tier Matcher

→

SQLite WAL Storage

Deterministic event clustering without heavy vector search dependencies. Guardrails prevent parameter mismatches (e.g. 2% vs 3% strikes).

Deliberately Lean: No microservices circus or cloud DBs. Concurrent HTTP/2 reads coupled with local SQLite WAL provided instant queries without network overhead.

Edge Runtimes & Distributed State

Cloudflare Edge AI Twin

</> Source

Open-source conversational twin on Workers and Durable Objects (<200ms latency) that lets recruiters vet my profile and skills. Replaced vector RAG with distilled in-memory context, cutting 200ms and stopping hallucinations; 4-tier model failover; privacy-preserving audit logs (salted SHA-256 IP hashing on D1).

Powers the live interactive widget on this site. Manages session lifecycles, per-IP rate limiting, and privacy-preserving audit logs via salted SHA-256 IP hashing on D1.

Latency Impact: -200ms TTFT reduction & zero factual hallucinations

The RAG Decision: Prototype ran classic Vectorize embeddings + top-k retrieval. Cosine similarity over 3 chunks missed basic facts and burned 2x inference quota. Replaced with distilled in-memory context: 100% accuracy, half the neuron burn.

Resilience: Automatically cascades across 4 models (Llama 3.1 8B → Llama 3.1 8B Fast → Llama 3.2 3B → Mistral 7B) to guarantee uptime during upstream provider rate limits.

Data Analysis & Econometrics

Zara Product Economics

</> Source

Audited 20,000+ SKUs with Negative Binomial regression, showed the public dataset is synthetic, and proved a ~59% unit lift still fails a net-margin test once discounts apply.

Refused the standard Kaggle ML story. Sales volume is artificially bounded (518–1,940), merchandising uses grocery labels on an apparel catalog, and sourcing-country means are indistinguishable from noise.

Audit: Synthetic DGP · promotion IRR 1.59 · lift fails a margin test

GLM, not vanity lift: Negative Binomial model on overdispersed counts. Promotion IRR 1.59 (p < 0.001). Log-price elastic. Near-shoring and product-copy terms insignificant. ANOVA on origin: p = 0.091.

Margin trap: A ~59% unit lift still loses ~20% gross profit under a 30% discount on 60% margin goods, before apparel return drag. The volume number does not survive contact with unit economics.

First Principles

Engineering Focus

⚡

Systems & Edge Runtimes

Building lean, low-latency applications that minimize unnecessary network hops. Stateful WebSockets, Durable Objects, lock-free SQLite WAL storage, and clean architectural boundaries.

🛡️

Vulnerability Discovery

Specialized focus on web application business logic flaws, parser normalization discrepancies, and state machine violations where automated scanners fail.

📊

Empirical Data Systems

Separating actual signal from noise in chaotic distributions. Event clustering, predictive market microstructure, risk constraints, and statistical retail analysis.

AppSec Practice

Security Research & Business Logic Flaws

Most severe modern web vulnerabilities aren't syntax typos—they are business logic flaws: assumptions about user sequence, state validation lapses, and discrepancies between what a reverse proxy accepts versus what a backend parser processes.

While engineering custom API pipelines, I uncovered and responsibly disclosed a critical business logic flaw at a major sports bookmaker that allowed covert, guaranteed payouts (~NGN 8M bounty awarded, ~$10,000 at July 2023 rates).

Responsible Disclosure

Parser Desync

State Machine Flaws

Let's talk engineering.

Open to discussions around systems engineering, application security, and high-signal data problems.

babalolabeat@gmail.com WhatsApp GitHub Profile Resume (PDF)