Who I Am
I operate at one standard — the one Netflix, Google, and Meta use when they ship ML to hundreds of millions of users.
I don't stop at model accuracy. I define SLOs before I write code. I measure latency at p50, p95, and p99. I write evaluation harnesses that catch regressions before they reach production. I document every failure in a postmortem log with a root cause and a fix.
Nine production-grade ML systems. Every one shipped with SLO definitions, ablation studies, latency benchmarks, quality gates, and documented production failures — not because a rubric asked for it, but because that's the only way to know if something actually works.
SLO-First
Define SLOs before writing code
Measured
Latency at p50, p95, and p99
Evaluated
Eval harnesses that catch regressions
Postmortems
Every failure root-caused and published
Education
Academic Background
Long Island University
🇺🇸 Brooklyn, New York, USA
MS in Artificial Intelligence
Panimalar Engineering College
🇮🇳 Chennai, Tamil Nadu, India
BE in Computer Science & Engineering
Technical Highlights
9 Production-Grade Systems
Scalable Personalization & Ranking
StreamLens
BM25 + FAISS + LightGBM LambdaRank · nDCG@10 = 0.7506 · +58% lift · p99 = 92ms · 1,000 concurrent users · 9 quality gates
Agentic RAG & LLM Systems
NeuraPilot
LangGraph 5-node DAG · RAGAS faithfulness = 0.81 · p95 = 4.2s · $0.00024/request · 8 bugs postmortem-documented
Security-Hardened AI Infrastructure
Voice RAG Assistant
RBAC at retrieval time · Recall@k = 1.000 · Security refusal = 1.000 (6/6) · $0.00/request — fully local
Real-Time ML & Streaming
GuardianDrive
Kafka + Airflow + Kubernetes · sub-second streaming inference · production safety gates
Clinical AI & Explainability
Health Risk Prediction
17 diseases · avg 87.3% · HIV 98.94% · TB 96.90% · SHAP-grounded interpretability
Computer Vision & Autonomous Perception
OpenDriveFM
6-camera BEV · Temporal Transformer · ADE = 2.457m · 18.4% over baseline · no pretrained weights
Cryptographic Systems
StegoCloak
AES-256-GCM · scrypt KDF · zero plaintext markers · 58 tests passing
Edge AI & Speech
KWS & Distress Detection
88.37% clean · 77.02% at 0dB SNR · 0.43 MB model size · p95 = 2.18ms
Financial Intelligence
Crypto Risk Bench
AAPL MAPE 3.34% · BTC MAPE 3.37% · 91% latency reduction via TTL cache
Volunteer Full-Stack Developer
- Co-architected and shipped a production-grade public website from zero — owning the full stack from information architecture and UI design through backend implementation — delivering a live, scalable web presence now serving as the organization's primary digital front door for thousands of community members. (Developer credit listed in footer: peopleinneed.org)
- Designed and engineered a unified backend operations system with local database storage powering four mission-critical nonprofit workflows — donation processing, volunteer intake, event registrations, and help request tracking — replacing fragmented manual processes and giving the internal team a single source of truth for community follow-up.
- Built and deployed a monthly newsletter pipeline automating subscriber communications end-to-end — covering event announcements, program recaps, and upcoming initiatives — eliminating a previously manual outreach bottleneck and enabling consistent, scalable engagement with the organization's growing member base.
Node.js
Python
Docker
Git
GitHub
FastAPISoftware Development Intern
- Owned end-to-end documentation for assigned software modules — translated ambiguous product requirements into precise algorithm designs and flowcharts, cutting documentation turnaround and establishing a reusable template adopted across the team's delivery workflow.
- Drove defect resolution across multiple software components by executing structured testing protocols, reproducing failure conditions with precision, and delivering root-cause analysis with actionable fix recommendations directly to the reporting manager — accelerating debugging cycles and reducing rework.
PythonGitGitHub
PytestCryptocurrency Price Prediction
AIP Conference Proceedings, 3175(1), 020004 · ICONIC 2K23 · 2025
Akila Lourdes Miriyala Francis (First Author), Esaiyarasi Viswanathan, Dheeksha Jayaraman, Janani Padamanavan Ashokkumar & Kavitha Subramani
As First Author, led end-to-end research, design, and writing of a peer-reviewed ML paper published in AIP Conference Proceedings — indexed across global academic databases. Applies LASSO regression to cryptocurrency price prediction, directly attacking the time complexity problem that plagues most existing Bitcoin forecasting models. By combining LASSO feature selection with ML-driven pattern recognition on historical market data, the system identifies stable, repeatable price trends with reduced computational overhead — demonstrating that intelligent feature engineering outperforms brute-force deep learning on high-noise financial signals.
Sign Language Recognition using CNN Algorithm with Machine Learning Techniques
IMPULSE 2K23 · Panimalar Engineering College, Chennai, India · 2023
M Akilan & M F Akila Lourdes
Researched, built, and presented a real-time sign language recognition system — combining CNNs for spatial gesture feature extraction with LSTMs for temporal sequence modeling. Translates hand gestures from the Indian Sign Language alphabet into text and audio output, addressing one of the most persistent accessibility barriers faced by 63 million deaf and hard-of-hearing individuals in India. Presented to an audience of researchers, faculty, and engineers — demonstrating both technical depth and the ability to communicate complex AI systems to a live expert audience.
IDEATHON 2K22 — 3rd Place Winner
Panimalar Engineering College, Chennai, India · 2022
Secured 3rd place in a college-wide innovation competition — where teams have limited time to conceive an original problem, design a viable solution, build a working prototype, and defend it in front of a technical judging panel. The win demanded full-stack thinking: rapid ideation, architecture design, technical implementation, and high-stakes presentation under pressure. Placing in the top 3 is a direct signal of the ability to execute original ideas fast — the core skill FAANG engineering internships demand from day one.
School Pupil Leader
Secondary School, India
Selected through a rigorous faculty and peer evaluation process to serve as School Pupil Leader — the highest student leadership position in the institution. Responsible for representing the entire student body, coordinating school-wide events, and serving as the primary communication bridge between students and faculty. Managed competing stakeholder needs, resolved student concerns through structured dialogue, and drove participation in academic and extracurricular initiatives. This role — earned before university, before any technical credential — is the earliest evidence of ownership, accountability, communication under pressure, and the trust of peers and authority figures.
Introduction to AI and Machine Learning on Google Cloud
Quantitative Research Job Simulation
Software Engineering Job Simulation
Contact Form
Please contact me directly at akilalourdes(at)gmail.com or drop your info here.