Skip to content

Optional analytics

Allow optional analytics to help improve projects, writing and navigation. It uses a pseudonymous first-party identifier after consent. No names, email addresses, messages or search text are collected.

Read the privacy policy
Allen Manoj
Work

The work behind the claims.

These are the builds I would show someone before asking them to trust me with theirs: what was frustrating, what I made, what exists now, and where the edges still are.

View selected systems Start a conversation
Current systems

Four problems I care about solving properly.

01 · Website intelligence system · In development

Lens

Turns website signals into a prioritised fix list.

Built around Crawl, Extract, Validate, Score, Explain, Monitor.

Replaces

Replaces vague website feedback with a scored audit and prioritised fixes.

Produces

Turns website signals into a prioritised fix list.

Evidence

Current product build · System architecture

View system notes Open current build
Diagnostic preview
site health
82
Truststrong
AI readabilityfix
Conversionfix
Example finding: make the service, evidence, and next action explicit.
02 · Account health and prioritisation system

Automated Intelligence Revenue System (AIRS)

Brings product, CRM, and billing signals into one ranked account view with the reasons visible.

Built around BigQuery, dbt, Features, XGBoost, SHAP, Airflow, Power BI.

Replaces

Helps revenue teams spot accounts losing momentum before the signal gets buried in exports and dashboards.

Produces

Brings product, CRM, and billing signals into one ranked account view with the reasons visible.

Evidence

Current system design

View system notes
Workflow map
product and account data
governed features
explainable score
revenue action
03Evidence-first content system · In development

DistributionOS

Turns trusted source material into grounded, reusable content without losing voice.

Editorial workspace
source library
notes03
research04
signals05
draft comparisonvoice rules
Replaces

Replaces scattered ideas and manual drafting with a source-grounded editorial workflow.

Produces

Turns trusted source material into grounded, reusable content without losing voice.

Evidence

Current system design

Built around Sources, Safe retrieval, Project context, Draft, Edit.

View system notes
04Internal accelerator · Current build

BrandScan

Turns a live website into reusable visual tokens and structured brand signals.

Token board
tokens.jsonbrand.css
Replaces

Replaces manual visual inspection with extracted colours, typography, contrast checks, and design tokens.

Produces

Turns a live website into reusable visual tokens and structured brand signals.

Evidence

Working website crawl · Extracted colour and typography results · Extracted component results · Download sample token output

Built around Website, Crawl, Extract, Structure, Store.

View system notes
Deployed engineering proof

Sentinel

Turns changing web information into evidence-backed alerts and briefings.

A deployed, open-source monitoring system that preserves source evidence before AI evaluates significance and routes an alert.

Replaces

Replaces noisy monitoring feeds with scored changes, source evidence, and useful escalation.

Produces

Turns changing web information into evidence-backed alerts and briefings.

Evidence

Open current build · View source

View Sentinel case study Open current build View source
Evidence feed
01source capturedok
02change scored91
03alert routedok
Earlier technical work

Personal systems first, followed by earlier analytics and research.

Kept compact for context. Open a group only when you want the detail.

Personal systemsSystems built for personal use - included because they show what gets built when there is no brief.3 items

LifeOS↗

Personal planning workspace with daily and seven-day planning, Google Calendar conflict detection, productivity analytics, push notification reminders, and a PWA wrapper for home screen use.

Next.js 16React 19ConvexClerkPWAWeb Push

Morsel↗

Health and recovery tracking PWA with natural language meal logging, workout volume trends, supplement tracking, recovery status, macro adherence, hydration analysis, and Claude-powered end-of-day reviews.

Next.jsFastAPISupabaseClaude APIRechartsPWA

Haven↗

Generative ambient audio engine for deep work. Uses Llama via Cloudflare AI Workers to design evolving sound textures, ElevenLabs for audio synthesis, Durable Objects for session state, R2 for audio caching, and a gapless Web Audio API player.

Next.jsCloudflare WorkersDurable ObjectsElevenLabsWeb Audio API
Machine learning & deep learning6 items

Histopathology Image Classification

SVM, MLP, and CNN models for nine-class colorectal cancer tissue classification. Reached 72% validation accuracy using stratified 5-fold cross-validation and Keras Tuner hyperparameter optimisation.

PythonTensorFlowscikit-learnKeras TunerCNN

Pulmonary Disease Classification↗

Deep learning framework for multiclass respiratory disease classification from audio. Raw recordings converted to log-mel spectrogram images and classified with CNN and EfficientNetB0 transfer learning across seven disease classes.

PythonCNNEfficientNetB0Deep LearningSignal Processing

Mental Health Treatment Prediction↗

Predicted adult mental health treatment outcomes from employee survey data. Compared Logistic Regression, Decision Trees, Random Forest, KNN, and SVM with feature selection and cross-validation. Best model reached 84.23% accuracy.

PythonFlaskscikit-learnPandasClassification

Crop Mapping - Fused Optical-Radar Classification

Seven-class crop type classification from fused optical and synthetic aperture radar satellite imagery. Optimised LDA, Logistic Regression, KNN, SVM, and Random Forest models with feature engineering on multi-temporal satellite bands.

RMachine LearningRemote SensingSARClassification

Age and Gender Detection↗

Real-time age and gender prediction from images and live webcam streams. CNN-based CaffeNet architecture with multi-output classification for age range and gender across multiple demographic groups.

Computer VisionCNNCaffeNetOpenCVPython

Mood-Based Music Recommender

Playlist recommender combining live facial emotion detection with DenseNet121 and text sentiment analysis with DistilBERT to match a user's current state to Spotify track features including valence, tempo, and energy.

DenseNet121DistilBERTStreamlitSpotify APIEmotion Detection
Data & analytics6 items

Image Captioning Web App

Production-grade Flask application on AWS with EC2 Auto Scaling, an Application Load Balancer, S3-triggered Lambda for async processing, EventBridge scheduling, RDS for persistence, Gemini caption generation, and CloudWatch observability.

FlaskEC2S3LambdaEventBridgeRDSCloudWatch

ETL Pipeline - AWS Data Infrastructure↗

End-to-end ETL pipeline for structured data transfer, transformation, error reporting, and dashboard delivery. Built on Redshift as the warehouse, Glue for orchestration, CloudWatch for error alerting, QuickSight for visualisation, and PostgreSQL for operational storage.

RedshiftAWS GlueCloudWatchQuickSightPostgreSQL

Churn Analysis Dashboard↗

Interactive Tableau dashboard covering churn patterns, KPIs, cohort analysis, churn prediction metrics, and retention segmentation. Includes executive summary view and drill-down by customer segment.

TableauData VisualisationCohort AnalysisRetention Analytics

Customer Segmentation

K-means clustering on customer behaviour and transaction data. Identified segments with different spending patterns and purchase frequency, used to inform targeted marketing strategy and retention priorities.

PythonPandasNumPyK-MeansMatplotlibSeaborn

Smart Lender - Loan Eligibility Prediction↗

Web application for loan eligibility prediction using XGBoost trained on applicant financial and demographic data. Includes risk scoring dashboards, eligibility decision logic, and an IBM Watson chatbot for applicant guidance.

PythonFlaskXGBoostIBM CloudIBM Watson

Project Management System - Bhumi

WhatsApp-based service request system for Bhumi's Tech4Change initiative. Users submit requests via WhatsApp and receive automated status updates through a Twilio integration, with a Streamlit admin interface for the operations team.

PythonTwilioStreamlitWhatsApp APIAutomation
Have a related system to build?

Bring the workflow, evidence, and decision that need to connect.

A rough description is enough to identify the useful boundary and a sensible first step.

Start a conversation