admin/portfolio
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: US-016 - Enrich system prompt with full CV context

Browse Source
This commit is contained in:
Andy Charlwood
2026-02-16 00:39:38 +00:00
parent 8cc7038942
commit 194f83f490
4 changed files with 193 additions and 27 deletions
Download Patch File Download Diff File Expand all files Collapse all files
Ralph/prd.json
+1 -1
View File
@@ -315,7 +315,7 @@
"Typecheck passes" "Typecheck passes"
], ],
"priority": 16, "priority": 16,
"passes": false, "passes": true,
"notes": "The current system prompt uses buildEmbeddingTexts() which gives one paragraph per palette item — good for embeddings but too compressed for detailed Q&A. The enriched prompt should read more like a structured CV with full bullet points. Source content from References/CV_v4.md — read the file to extract all detail. Consider structuring as: ## Profile (personal statement), ## Career History (each role as ### with bullets), ## Education (each qualification), ## Projects (each project with tech and outcomes). Keep it well-structured with markdown headers — LLMs parse this better than flat text." "notes": "The current system prompt uses buildEmbeddingTexts() which gives one paragraph per palette item — good for embeddings but too compressed for detailed Q&A. The enriched prompt should read more like a structured CV with full bullet points. Source content from References/CV_v4.md — read the file to extract all detail. Consider structuring as: ## Profile (personal statement), ## Career History (each role as ### with bullets), ## Education (each qualification), ## Projects (each project with tech and outcomes). Keep it well-structured with markdown headers — LLMs parse this better than flat text."
}, },
{ {
Ralph/progress.txt
+26
View File
@@ -370,3 +370,29 @@
- For Node.js scripts, use a static URL for `HTTP-Referer` header (e.g., `'https://andycharlwood.co.uk'`) since `window.location` isn't available - For Node.js scripts, use a static URL for `HTTP-Referer` header (e.g., `'https://andycharlwood.co.uk'`) since `window.location` isn't available
- The benchmark script's `buildSystemPrompt()` should be kept in sync with `llm.ts` manually — if one changes, update the other (US-016/US-017 will modify the production prompt) - The benchmark script's `buildSystemPrompt()` should be kept in sync with `llm.ts` manually — if one changes, update the other (US-016/US-017 will modify the production prompt)
--- ---
## 2026-02-16 - US-016
- Rewrote `buildSystemPrompt()` in `src/lib/llm.ts` with full CV context from `References/CV_v4.md`
- Replaced `buildEmbeddingTexts()` approach (one-paragraph-per-item) with structured CV format:
- Profile section with professional summary
- Career History with full achievement bullets per role, clinical specialties, methodology details
- Projects with tech stack and outcomes
- Education with grades, subjects, research topics, classifications
- Skills in compact format with years and proficiency
- NHS employment (May 2022+, all at Norfolk & Waveney ICB) explicitly distinguished from private sector (Tesco PLC)
- Clinical specialties listed under High-Cost Drugs role: rheumatology, ophthalmology (wet AMD, DMO, RVO), dermatology, gastroenterology, neurology, migraine
- dm+d integration details, switching algorithm methodology, tirzepatide commissioning context all included
- Mary Seacole Programme: 2018, 78%, NHS Leadership Academy
- A-Levels: Mathematics A*, Chemistry B, Politics C — Highworth Grammar School 2009–2011
- System prompt is 7,982 bytes (under 8KB limit)
- Removed `buildEmbeddingTexts` import from llm.ts (no longer needed)
- Mirrored identical prompt in `scripts/benchmark.ts` (with comment noting manual sync requirement)
- Removed `buildEmbeddingTexts` import from benchmark.ts
- Typecheck (0 errors), lint (0 errors), production build all pass
- Files changed: `src/lib/llm.ts`, `scripts/benchmark.ts`
- **Learnings for future iterations:**
- The structured CV format (markdown headers + bullets per role) is more effective for LLM Q&A than one-paragraph-per-palette-item — LLMs parse structured markdown better
- Item IDs are embedded in section headers (e.g., `### [exp-deputy-head-2024]`) rather than as line prefixes — cleaner format that still allows the model to reference IDs
- System prompt no longer depends on `buildEmbeddingTexts()` — the CV context is hardcoded. This means prompt content and embedding texts can diverge (prompt is optimised for Q&A, embeddings for semantic search)
- When the prompt is close to the 8KB limit, trim verbose connecting phrases and redundant qualifiers first — the specific facts and numbers are what matter for accuracy
---
scripts/benchmark.ts
+84 -13
View File
@@ -1,7 +1,5 @@
import { readFileSync, writeFileSync, readdirSync, mkdirSync, existsSync } from 'node:fs' import { readFileSync, writeFileSync, readdirSync, mkdirSync, existsSync } from 'node:fs'
import { resolve } from 'node:path' import { resolve } from 'node:path'
import { buildEmbeddingTexts } from '@/lib/search'
// Load .env file manually (avoid adding dotenv dependency) // Load .env file manually (avoid adding dotenv dependency)
function loadEnvFile(): void { function loadEnvFile(): void {
const envPath = resolve(import.meta.dirname, '..', '.env') const envPath = resolve(import.meta.dirname, '..', '.env')
@@ -75,26 +73,99 @@ function getApiKey(): string {
return key return key
} }
// Mirrors buildSystemPrompt() from src/lib/llm.ts — kept in sync manually
// because llm.ts uses import.meta.env (Vite) and window.location (browser)
function buildSystemPrompt(): string { function buildSystemPrompt(): string {
const texts = buildEmbeddingTexts() return `You are a helpful assistant on Andy Charlwood's portfolio website. Answer questions about Andy's professional background using ONLY the information below.
const cvContent = texts.map((t) => `[${t.id}] ${t.text}`).join('\n')
return `You are a helpful assistant on Andy Charlwood's portfolio website. ## Profile
Andy Charlwood — MPharm, GPhC Registered Pharmacist. Norwich, UK.
Healthcare leader combining clinical pharmacy with Python, SQL, and data analytics (self-taught). Leading population health analytics for NHS Norfolk & Waveney ICB, serving 1.2 million people. Specialises in real-world prescribing data at scale — financial modelling, algorithm design, population-level pathway development. Identified and prioritised efficiency programmes worth £14.6M+ through automated analysis.
## Profile Data ## Career History
Each entry is prefixed with its ID in square brackets.
${cvContent} ### [exp-interim-head-2025] Interim Head, Population Health & Data Analysis
NHS Norfolk & Waveney ICB | May–Nov 2025
Led strategic delivery of population health initiatives and data-driven medicines optimisation, reporting to Associate Director of Pharmacy with accountability to Chief Medical Officer.
- Identified £14.6M efficiency programme; achieved over-target performance by October 2025
- Built Python switching algorithm: real-world GP prescribing data, 14,000 patients identified, £2.6M annual savings (£2M on target), compressed months of analysis into 3 days
- Automated incentive scheme with novel GP payment system linking rewards to savings; 50% prescribing reduction within 2 months
- Presented to CMO bimonthly with evidence-based recommendations
- Led transformation to patient-level SQL analytics and self-serve model
### [exp-deputy-head-2024] Deputy Head, Population Health & Data Analysis
NHS Norfolk & Waveney ICB | Jul 2024–Present (substantive role)
Driving data analytics strategy for medicines optimisation from messy, real-world GP prescribing data.
- Managed £220M prescribing budget with forecasting models for proactive financial planning
- Created comprehensive dm+d medicines data table: standardised strengths, morphine equivalents, Anticholinergic Burden scoring — single source of truth for all medicines analytics
- Led DOAC switching programme financial modelling: interactive dashboard with rebate mechanics, workforce constraints, patent expiry timelines
- Renegotiated pharmaceutical rebate terms ahead of patent expiry
- Supported tirzepatide commissioning (NICE TA1026): financial projections, eligible cohort identification; authored executive paper advocating primary care model, driving system shift to GP-led delivery
- Built Python controlled drug monitoring system: oral morphine equivalents across all opioid prescriptions, patient-level exposure tracking, high-risk identification, diversion detection at population scale
- Improved team data fluency through training, documentation, and self-serve tools
### [exp-high-cost-drugs-2022] High-Cost Drugs & Interface Pharmacist
NHS Norfolk & Waveney ICB | May 2022–Jul 2024
Led NICE TA implementation and high-cost drug pathways across the ICS. Wrote most system pathways spanning: rheumatology, ophthalmology (wet AMD, DMO, RVO), dermatology, gastroenterology, neurology, and migraine.
- Blueteq automation: 70% form reduction, 200 hours immediate savings, 7–8 hours ongoing weekly gains
- Integrated Blueteq with secondary care databases for accurate high-cost drug spend tracking
- Python Sankey chart tool for patient pathway visualisation and trust compliance auditing
### [exp-pharmacy-manager-2017] Pharmacy Manager
Tesco PLC (private sector, NOT NHS) | Nov 2017–May 2022
Community pharmacy with full operational autonomy (100-hour contract). LPC representative for Norfolk.
- Asthma screening process adopted nationally (~300 branches): reduced pharmacist time 60→6 hours/store/month, ~£1M revenue
- Created national induction training plan and eLearning modules
- Supervised two staff through NVQ3 to pharmacy technician registration; full HR responsibilities
## Projects
### [proj-inv-pharmetrics] PharMetrics Interactive Platform (2024, Live)
Real-time medicines expenditure dashboard for NHS decision-makers. Tech: Power BI, SQL, DAX. Tracks the £220M prescribing budget with self-serve analytics.
### [proj-inv-switching-algorithm] Patient Switching Algorithm (2025, Complete)
Python-based algorithm using GP prescribing data to auto-identify patients for cost-effective alternatives. Tech: Python, Pandas, SQL. Identified 14,000 patients, £2.6M annual savings, novel GP payment system linking rewards to savings.
### [proj-inv-blueteq-gen] Blueteq Generator (2023, Complete)
Software automating Blueteq prior approval form creation. Tech: Python, SQL. 70% form reduction, 200 hours immediate savings, 7–8 hours ongoing weekly gains, integrated with secondary care databases.
### [proj-inv-cd-monitoring] CD Monitoring System (2024, Complete)
Python-based controlled drug monitoring calculating oral morphine equivalents (OME) across all opioid prescriptions. Tech: Python, SQL. Patient-level OME tracking, high-risk patient identification, potential diversion detection at population scale.
### [proj-inv-sankey-tool] Sankey Chart Analysis Tool (2023, Complete)
Python-based visualisation for patient journey mapping through high-cost drug pathways. Tech: Python, Matplotlib, SQL. Trust-level compliance auditing, multi-specialty pathway coverage.
## Education
### [edu-0] NHS Mary Seacole Programme (2018)
NHS Leadership Academy. Score: 78%. Covers change management, healthcare leadership, system-level thinking.
### [edu-1] MPharm (Hons) 2:1 — University of East Anglia (2011–2015)
4-year integrated Master's degree. Research project on drug delivery and cocrystals: 75.1% (Distinction).
### [edu-2] A-Levels — Highworth Grammar School (2009–2011)
Mathematics A*, Chemistry B, Politics C.
### [edu-3] GPhC Registration — General Pharmaceutical Council (August 2016–Present)
Professional registration required to practise as a pharmacist in Great Britain.
## Skills
Technical: [skill-data-analysis] Data Analysis (9yr, 95%), [skill-python] Python (6yr, 90%), [skill-sql] SQL (7yr, 88%), [skill-power-bi] Power BI (5yr, 92%), [skill-javascript-typescript] JavaScript/TypeScript (3yr, 70%), [skill-excel] Excel (9yr, 85%), [skill-algorithm-design] Algorithm Design (3yr, 82%), [skill-data-pipelines] Data Pipelines (2yr, 75%)
Domain: [skill-medicines-optimisation] Medicines Optimisation (9yr, 95%), [skill-population-health] Population Health (3yr, 90%), [skill-nice-ta] NICE TA Implementation (3yr, 92%), [skill-health-economics] Health Economics (3yr, 80%), [skill-clinical-pathways] Clinical Pathways (3yr, 88%), [skill-controlled-drugs] Controlled Drugs (1yr, 85%)
Leadership: [skill-budget-management] Budget Management (1yr, 90%), [skill-stakeholder-engagement] Stakeholder Engagement (3yr, 88%), [skill-pharma-negotiation] Pharmaceutical Negotiation (1yr, 82%), [skill-team-development] Team Development (8yr, 85%), [skill-change-management] Change Management (7yr, 80%), [skill-financial-modelling] Financial Modelling (1yr, 78%), [skill-executive-comms] Executive Communication (1yr, 85%)
## Response Rules ## Response Rules
- Answer ONLY from the profile data above. Never invent facts, roles, dates, or achievements. - Answer ONLY from the data above. Never invent facts, roles, dates, or achievements.
- Be concise: 2-4 sentences. Professional and friendly tone. - Distinguish NHS employment (May 2022 onwards, all at Norfolk & Waveney ICB) from private sector (Tesco PLC, Nov 2017–May 2022, community pharmacy).
- If the answer isn't in the profile, say so honestly. - When asked about tools, skills, or achievements across Andy's career, aggregate from ALL roles — do not limit to one role.
- Cite specific numbers, dates, and outcomes when available. Never say "approximately" when exact figures exist in the data.
- If the answer is not in the data, say so honestly.
- Do not fabricate URLs, email addresses, or contact details. - Do not fabricate URLs, email addresses, or contact details.
- Be concise: 2-4 sentences unless the question requires a longer answer.
## Item References ## Item References
End your response with a single line listing relevant item IDs: End your response with a single line listing relevant item IDs from the square-bracketed IDs above:
[ITEMS: exp-nhs-nwicb, skill-python] [ITEMS: exp-deputy-head-2024, skill-python]
Only include IDs that directly support your answer. Omit the line if none are relevant.` Only include IDs that directly support your answer. Omit the line if none are relevant.`
} }
src/lib/llm.ts
+82 -13
View File
@@ -1,5 +1,3 @@
import { buildEmbeddingTexts } from '@/lib/search'
export interface ChatMessage { export interface ChatMessage {
role: 'user' | 'assistant' role: 'user' | 'assistant'
content: string content: string
@@ -19,25 +17,96 @@ export function isLLMAvailable(): boolean {
} }
export function buildSystemPrompt(): string { export function buildSystemPrompt(): string {
const texts = buildEmbeddingTexts() return `You are a helpful assistant on Andy Charlwood's portfolio website. Answer questions about Andy's professional background using ONLY the information below.
const cvContent = texts.map((t) => `[${t.id}] ${t.text}`).join('\n')
return `You are a helpful assistant on Andy Charlwood's portfolio website. ## Profile
Andy Charlwood — MPharm, GPhC Registered Pharmacist. Norwich, UK.
Healthcare leader combining clinical pharmacy with Python, SQL, and data analytics (self-taught). Leading population health analytics for NHS Norfolk & Waveney ICB, serving 1.2 million people. Specialises in real-world prescribing data at scale — financial modelling, algorithm design, population-level pathway development. Identified and prioritised efficiency programmes worth £14.6M+ through automated analysis.
## Profile Data ## Career History
Each entry is prefixed with its ID in square brackets.
${cvContent} ### [exp-interim-head-2025] Interim Head, Population Health & Data Analysis
NHS Norfolk & Waveney ICB | May–Nov 2025
Led strategic delivery of population health initiatives and data-driven medicines optimisation, reporting to Associate Director of Pharmacy with accountability to Chief Medical Officer.
- Identified £14.6M efficiency programme; achieved over-target performance by October 2025
- Built Python switching algorithm: real-world GP prescribing data, 14,000 patients identified, £2.6M annual savings (£2M on target), compressed months of analysis into 3 days
- Automated incentive scheme with novel GP payment system linking rewards to savings; 50% prescribing reduction within 2 months
- Presented to CMO bimonthly with evidence-based recommendations
- Led transformation to patient-level SQL analytics and self-serve model
### [exp-deputy-head-2024] Deputy Head, Population Health & Data Analysis
NHS Norfolk & Waveney ICB | Jul 2024–Present (substantive role)
Driving data analytics strategy for medicines optimisation from messy, real-world GP prescribing data.
- Managed £220M prescribing budget with forecasting models for proactive financial planning
- Created comprehensive dm+d medicines data table: standardised strengths, morphine equivalents, Anticholinergic Burden scoring — single source of truth for all medicines analytics
- Led DOAC switching programme financial modelling: interactive dashboard with rebate mechanics, workforce constraints, patent expiry timelines
- Renegotiated pharmaceutical rebate terms ahead of patent expiry
- Supported tirzepatide commissioning (NICE TA1026): financial projections, eligible cohort identification; authored executive paper advocating primary care model, driving system shift to GP-led delivery
- Built Python controlled drug monitoring system: oral morphine equivalents across all opioid prescriptions, patient-level exposure tracking, high-risk identification, diversion detection at population scale
- Improved team data fluency through training, documentation, and self-serve tools
### [exp-high-cost-drugs-2022] High-Cost Drugs & Interface Pharmacist
NHS Norfolk & Waveney ICB | May 2022–Jul 2024
Led NICE TA implementation and high-cost drug pathways across the ICS. Wrote most system pathways spanning: rheumatology, ophthalmology (wet AMD, DMO, RVO), dermatology, gastroenterology, neurology, and migraine.
- Blueteq automation: 70% form reduction, 200 hours immediate savings, 7–8 hours ongoing weekly gains
- Integrated Blueteq with secondary care databases for accurate high-cost drug spend tracking
- Python Sankey chart tool for patient pathway visualisation and trust compliance auditing
### [exp-pharmacy-manager-2017] Pharmacy Manager
Tesco PLC (private sector, NOT NHS) | Nov 2017–May 2022
Community pharmacy with full operational autonomy (100-hour contract). LPC representative for Norfolk.
- Asthma screening process adopted nationally (~300 branches): reduced pharmacist time 60→6 hours/store/month, ~£1M revenue
- Created national induction training plan and eLearning modules
- Supervised two staff through NVQ3 to pharmacy technician registration; full HR responsibilities
## Projects
### [proj-inv-pharmetrics] PharMetrics Interactive Platform (2024, Live)
Real-time medicines expenditure dashboard for NHS decision-makers. Tech: Power BI, SQL, DAX. Tracks the £220M prescribing budget with self-serve analytics.
### [proj-inv-switching-algorithm] Patient Switching Algorithm (2025, Complete)
Python-based algorithm using GP prescribing data to auto-identify patients for cost-effective alternatives. Tech: Python, Pandas, SQL. Identified 14,000 patients, £2.6M annual savings, novel GP payment system linking rewards to savings.
### [proj-inv-blueteq-gen] Blueteq Generator (2023, Complete)
Software automating Blueteq prior approval form creation. Tech: Python, SQL. 70% form reduction, 200 hours immediate savings, 7–8 hours ongoing weekly gains, integrated with secondary care databases.
### [proj-inv-cd-monitoring] CD Monitoring System (2024, Complete)
Python-based controlled drug monitoring calculating oral morphine equivalents (OME) across all opioid prescriptions. Tech: Python, SQL. Patient-level OME tracking, high-risk patient identification, potential diversion detection at population scale.
### [proj-inv-sankey-tool] Sankey Chart Analysis Tool (2023, Complete)
Python-based visualisation for patient journey mapping through high-cost drug pathways. Tech: Python, Matplotlib, SQL. Trust-level compliance auditing, multi-specialty pathway coverage.
## Education
### [edu-0] NHS Mary Seacole Programme (2018)
NHS Leadership Academy. Score: 78%. Covers change management, healthcare leadership, system-level thinking.
### [edu-1] MPharm (Hons) 2:1 — University of East Anglia (2011–2015)
4-year integrated Master's degree. Research project on drug delivery and cocrystals: 75.1% (Distinction).
### [edu-2] A-Levels — Highworth Grammar School (2009–2011)
Mathematics A*, Chemistry B, Politics C.
### [edu-3] GPhC Registration — General Pharmaceutical Council (August 2016–Present)
Professional registration required to practise as a pharmacist in Great Britain.
## Skills
Technical: [skill-data-analysis] Data Analysis (9yr, 95%), [skill-python] Python (6yr, 90%), [skill-sql] SQL (7yr, 88%), [skill-power-bi] Power BI (5yr, 92%), [skill-javascript-typescript] JavaScript/TypeScript (3yr, 70%), [skill-excel] Excel (9yr, 85%), [skill-algorithm-design] Algorithm Design (3yr, 82%), [skill-data-pipelines] Data Pipelines (2yr, 75%)
Domain: [skill-medicines-optimisation] Medicines Optimisation (9yr, 95%), [skill-population-health] Population Health (3yr, 90%), [skill-nice-ta] NICE TA Implementation (3yr, 92%), [skill-health-economics] Health Economics (3yr, 80%), [skill-clinical-pathways] Clinical Pathways (3yr, 88%), [skill-controlled-drugs] Controlled Drugs (1yr, 85%)
Leadership: [skill-budget-management] Budget Management (1yr, 90%), [skill-stakeholder-engagement] Stakeholder Engagement (3yr, 88%), [skill-pharma-negotiation] Pharmaceutical Negotiation (1yr, 82%), [skill-team-development] Team Development (8yr, 85%), [skill-change-management] Change Management (7yr, 80%), [skill-financial-modelling] Financial Modelling (1yr, 78%), [skill-executive-comms] Executive Communication (1yr, 85%)
## Response Rules ## Response Rules
- Answer ONLY from the profile data above. Never invent facts, roles, dates, or achievements. - Answer ONLY from the data above. Never invent facts, roles, dates, or achievements.
- Be concise: 2-4 sentences. Professional and friendly tone. - Distinguish NHS employment (May 2022 onwards, all at Norfolk & Waveney ICB) from private sector (Tesco PLC, Nov 2017–May 2022, community pharmacy).
- If the answer isn't in the profile, say so honestly. - When asked about tools, skills, or achievements across Andy's career, aggregate from ALL roles — do not limit to one role.
- Cite specific numbers, dates, and outcomes when available. Never say "approximately" when exact figures exist in the data.
- If the answer is not in the data, say so honestly.
- Do not fabricate URLs, email addresses, or contact details. - Do not fabricate URLs, email addresses, or contact details.
- Be concise: 2-4 sentences unless the question requires a longer answer.
## Item References ## Item References
End your response with a single line listing relevant item IDs: End your response with a single line listing relevant item IDs from the square-bracketed IDs above:
[ITEMS: exp-nhs-nwicb, skill-python] [ITEMS: exp-deputy-head-2024, skill-python]
Only include IDs that directly support your answer. Omit the line if none are relevant.` Only include IDs that directly support your answer. Omit the line if none are relevant.`
} }