State-level snapshot · how Uttarakhand's health system is shaped
| District | SC | PHC | CHC | SDH | DH | Total |
|---|---|---|---|---|---|---|
| State total | 1,765 | 239 | 55 | 20 | 18 | 2,097 |
Each district is administratively split into development blocks, the unit at which Block Health Societies operate and BMOICs sit. Pauri Garhwal, the largest district by area, runs 15 blocks; Rudraprayag and Bageshwar manage with 3 each.
Annual snapshots of facility counts and AAM rollout from FY 2018-19 to FY 2023-24. The dominant trend is AAM growth — from zero in 2018 to 2,186 operational by August 2024 — alongside gradual sub-centre consolidation.
DATA.timeSeries.Comparison with peer hill states — same 90:10 funding ratio, similar terrain challenges — plus the all-India per-crore-population baseline. Numbers normalised to population so smaller states (Sikkim) aren’t penalised.
DATA.hillStates.Inputs (facilities, money, AAMs) finally meet outputs. Every facility-tab in this document tells you what was built. This exhibit tells you whether it worked. Indicators are NFHS-4 (2015-16) vs NFHS-5 (2019-21) — the only two rounds available; NFHS-6 fieldwork is ongoing.
NFHS-5 fieldwork in Uttarakhand was completed by March 2021. That means it predates most of the AAM rollout (which scaled from 892 operational in FY 2020-21 to 2,186 in FY 2023-24). So the gains shown below come from the pre-AAM NHM architecture: facility upgrades, JSY/JSSK incentives, Mission Indradhanush, the ASHA cadre, and SUMAN. The next NFHS round will tell us whether the AAM scale-up moved the needle further.
DATA.outcomes.state.District-level drill-down · profiles, blocks, and side-by-side comparison
A consolidated card per district pulling together what's scattered across the map, hierarchy explorer, and block deep-dive. Click any card to expand its full block list, AAM rollout share and notable facilities.
districts array + DATA.aam.byDistrict + blocksByDistrict + DATA.namedFacilities.Pick any two of Uttarakhand’s 13 districts and see their facility counts and population catchment compared metric-by-metric. Used to answer questions like “is Almora better staffed than Pithoragarh?” in one glance.
Norms require a denominator. This exhibit shows, for each district, the 2024 projected population, the facilities actually present, and the facilities IPHS 2022 says should be there. The gap is the hard reality.
India's last completed census was 2011. The 2021 census was postponed (COVID); the next will be conducted in 2027 (phase 1 in snow-bound regions including Uttarakhand starts October 2026). Every "current" population in this file is a projection — the Technical Group on Population Projections (MoHFW) puts Uttarakhand at 11.8 million (118 lakh) as of July 2024. Plain districts (Dehradun, Haridwar, USN) have grown ~22% since 2011; hill districts (Pauri, Almora, Tehri) have lost population to out-migration. Where the gap looks "satisfied" in a hill district, it's often because the population is shrinking faster than facilities can be wound down — not because care is plentiful.
districts[].pop2024est + IPHS 2022 hill/plain ratios · norms: 1 SC per 3,000 (hill) / 5,000 (plain) · 1 PHC per 20,000 / 30,000 · 1 CHC per 80,000 / 120,000.For each district: NFHS-5 (2019-21) health outcomes alongside the SC/PHC/CHC density that produced them. Sort by any column to find productive districts (high outcomes, modest inputs) and under-performing ones (rich infrastructure, weaker results).
Colour coding marks each value against the state benchmark: green = ≥ 3 percentage points above state average, amber = within ±3pp, brick = ≥ 3pp below. The "SCs per 10k pop" column is the input density — the denominator any outcome gets normalised against. A district with high outcomes and low SC density is doing more with less; the reverse pattern flags a productivity gap.
DATA.outcomes.byDistrict.Services & people · what NHM delivers and through whom
The community-level health workforce in Uttarakhand stacks five tiers deep, from ASHAs in every village up to a small State team coordinating block-and-district mobilizers.
The state runs two parallel directorates — the regular allopathic line under the Department of Medical Health & Family Welfare, and the Directorate of Ayurvedic & Unani Services for AYUSH systems (Ayurveda, Yoga & Naturopathy, Unani, Siddha, Homeopathy). They converge inside NHM facilities through co-located AYUSH wings and through joint funding of medicines and salaries.
How the system is funded, planned, and where its institutions sit
Every named facility in the state catalogue — AIIMS, GMCs, district hospitals, key SDHs and major private institutions — with bed counts, ABDM ownership classification, and HFR registry status. Filter by district, type, ownership, or search by name. Each entry is structured to accept its HFR ID from the public registry at facility.abdm.gov.in.
DATA.namedFacilities · 30 records as of August 2024. ABDM ownership and facility-type classifications are derived from public knowledge; HFR IDs pending verification at facility.abdm.gov.in.Enter any population and terrain class; the calculator returns the number of Sub-centres, PHCs, CHCs and ASHAs that the area should have under Indian Public Health Standards 2022. A planning tool, not a reporting tool — useful for advocacy and scenario work.
DATA.norms in the catalog.From inputs to results · what the system actually delivers to women and children. NFHS-5 (2019-21), SRS, and the trend over four years.
Six core indicators from the National Family Health Survey round 5 (fieldwork Jan 2020 – Mar 2021), broken out by all 13 districts. Cell colour reflects performance against the state average — deficit (red) when more than 5pp below state, surplus (green) when more than 5pp above.
State-level progress between the two most recent national surveys. Each indicator shows the NFHS-4 baseline (2015-16) and NFHS-5 value (2019-21). Direction badge indicates whether the change is favourable.
Sample Registration System (SRS) data over 12 years. IMR has dropped from 38 to 22 per 1,000 live births — well below the national average of 28. Life expectancy at birth has improved from 70.1 to 72.5 years.
Each district is one dot. X-axis: sub-centres per 10,000 people (an input). Y-axis: institutional delivery rate (an outcome). If facility density caused better outcomes, you would expect a clear upward slope. What actually shows up is more complicated.
Frontispiece · Interactive district map at the top of this document.
A consolidated overview of the National Health Mission in Uttarakhand — its origin in 2005, the urban sub-mission added in 2013, the Health & Wellness Centre rollout under Ayushman Bharat, and the geographical asymmetries that make the state's structure distinct from the national template.
The National Rural Health Mission was launched in Uttarakhand on 27 October 2005 by the Union Cabinet Health Minister. Twenty years on, NHM does not replace the state's Department of Medical, Health & Family Welfare; it runs alongside it — the IAS Mission Director, the State Programme Management Unit's professional managers, and an army of contractually-engaged community workers operating through the same Chief Medical Officer who heads the regular district cadre.
The Union Cabinet's 2013 decision to make the National Urban Health Mission (NUHM) a sub-mission under the umbrella NHM — alongside the original NRHM — means today's structure covers both rural mountain panchayats and urban slums in Dehradun, Haridwar, Roorkee, Haldwani, and Rudrapur under a single roof.
What makes Uttarakhand unusual is geography. Nine of thirteen districts are classified hill districts. The Indian Public Health Standards apply different population norms here: a sub-centre serves 3,000 people in hills against 5,000 in plains, a PHC serves 20,000 against 30,000, and a CHC serves 80,000 against 120,000. Pauri Garhwal carries 218 sub-centres for an estimated population of 625,000; plain-state Haridwar runs 159 for nearly two million. The pages below trace this system from the Mission Director's office at Danda Lakhond down to the village ASHA — with the names, the numbers, and the maps that connect them.
As published on the official directory (last updated 31 December 2024). A three-headed apex with portfolios sliced across seven Assistant Directors. All sit at 3rd Floor, Directorate of Health, Danda Lakhond, Sahastradhara Road, Dehradun — on a single landline: 0135-2607938.
The IAS Mission Director provides administrative authority and leads UKHFWS (the Health & Family Welfare Society). The Director NHM is the senior medical-public-health professional head. The State Programme Manager runs the SPMU's day-to-day operational machinery. Below them, seven Assistant Directors carry portfolio responsibility across the full programmatic line.
Portfolio responsibility across the National Health Mission's full programmatic line. Read horizontally to see how the alphabet soup divides up.
All five tiers consolidated into a single view: state apex, district command, block management, four facility streams, and the community workforce that ties the system to villages. Cadre type is encoded by node colour.
As listed on the official NHM Uttarakhand directory. Each CMO holds dual authority: head of the regular Department of Medical, Health & Family Welfare in the district, and District Mission Director for NHM through the District Health Society.
A district-by-district inventory of sub-centres, Primary Health Centres, Community Health Centres, Sub-Divisional Hospitals, and District Hospitals across the state. Numbers from the Rural Health Statistics district-wise report.
Source: Rural Health Statistics, MoHFW.
| District | Terrain | SCs | PHCs | CHCs | SDH | DH | Total |
|---|
Dedicated women's hospitals carry the bulk of institutional deliveries in Uttarakhand. They operate as 24×7 First Referral Units offering OBG and paediatric services, free institutional delivery under JSY/JSSK, comprehensive abortion care, and post-natal care.
Eight districts have a dedicated District Female Hospital (DFH) listed under NHM's 24×7 facility roster. The remaining five rely on the Joint / District Hospital and a Combined Hospital with a women's wing. Per the CAG Performance Audit on District Hospital Outcomes (2021), SNCU services were available only at DFH Haridwar among the test-checked DFHs — a stark indicator of the staffing-and-equipment unevenness across districts.
Bageshwar, Champawat, Rudraprayag, Tehri Garhwal and Uttarkashi do not run a separate DFH. Maternal services are delivered through the District Hospital and 24×7 CHC/PHCs.
From the IGNOU-trained Community Health Officer at a Health & Wellness Centre, to the ASHA carrying a drug kit through monsoon-blocked panchayats, to the SPMU consultant tracking HMIS dashboards in Dehradun — the workforce that runs NHM in this state, with sanctioned numbers, qualifications, and pay where published.
Selected by Gram Sabha. Resident of the village. Carries a drug kit (DDK, paracetamol, IFA, ORS, povidone iodine, cotrimoxazole syrup, zinc, bandages). Refilled at PHC/CHC.
Pay: GoI activity-based incentives plus Uttarakhand fixed honorarium of ₹3,000/month plus 10% performance-based top-up.
Tools: uniform, diary, CUG SIM card. Reports to ANM (functionally) and ASHA Facilitator (administratively).
Heads each Health & Wellness Centre under Ayushman Bharat. Delivers Comprehensive Primary Health Care (CPHC) — the 12-service package including NCD screening, mental health, palliative care.
Qualifications: B.Sc Nursing, Post-Basic B.Sc Nursing, GNM with IGNOU 6-month CCH (Certificate in Community Health) bridge, or AYUSH practitioners with the bridge.
Recruitment: via HNB Uttarakhand Medical Education University. Walk-in / OMR-based exam followed by district-wise counselling.
ANM (Auxiliary Nurse Midwife) — the bridge between ASHA and the formal system. INC-recognised ANM training, registration with Uttarakhand Nursing Council. Conducts SBA-attended deliveries, ANC, immunisation.
MPW (Male) — multipurpose worker for surveillance and male health.
Staff Nurse / LHV — PHC and CHC level; trained in SBA and increasingly in cervical cancer screening (VIA) and OVE for oral cancer.
Walk-in interviews held monthly. Each candidate quotes their own monthly figure for a named hill posting; the empanelment committee accepts or counter-offers on the spot.
Recent rates: hill specialists ₹2.0–4.0 lakh/month (OBG, Paediatrics, Anaesthesia, Surgery); MBBS MOs ₹1.0–1.5 lakh/month. Plain districts cluster around ₹70,000–90,000.
Plus: the state's Durgam ("difficult-area") allowance for regular cadre posted to specified hill blocks.
SPMU: SPM, SAM, SDM, ~30 PMCs across MCH, RBSK, RKSK, NUHM, NCD, QA/QM, NTCP, PC-PNDT, Drug Warehousing, Civil Wing, MMU/RT, Community Process, Documentation, Biomedical Engineering, IT.
DPMU per district: District Programme Manager, Accounts Manager, Data Manager / M&E Officer, District Community Mobilizer, DEO + programme consultants.
BPMU per block: Block Programme Manager, Block Accountant, Block Data Manager, Block Community Mobilizer.
MMU teams: Mobile Medical Units (some run with The HANS Foundation) bring MO + nurse + lab tech + driver to underserved villages. Pharmacy-on-wheels concept under Mukhyamantri Chal Chikitsa Vahan Yojana.
108: EMRI-operated emergency ambulance with EMTs trained in BLS/ALS.
104: Integrated Health Helpline — under Dr. Aakanksha Nirala's portfolio — tele-triage, drug information, grievance redressal.
Nine hill districts and four plain districts run on the same NHM rules but with different IPHS norms, different cadre supply problems, and different referral logistics. Here is the contrast in numbers.
Indicative ranges from recent NHM Uttarakhand walk-in interview empanelments. The Durgam allowance is a state-paid hard-area allowance on top of basic pay for regular cadre posted to specified hill blocks.
↑ Return to contentsBetween April and November, an estimated 50–60 lakh transient pilgrims travel through Uttarkashi, Rudraprayag, Chamoli and Pithoragarh — districts whose otherwise-modest permanent population has no Sub-Divisional Hospital between them. NHM responds with a seasonal surge of contractual personnel and equipment.
Uttarakhand runs two parallel medical directorates. The regular allopathic line under the Department of Medical Health & Family Welfare, and the Directorate of Ayurvedic & Unani Services for AYUSH systems. They converge inside NHM facilities through co-located AYUSH wings, and they share staffing through the Community Health Officer bridge.
The AYUSH directorate is headquartered in Dehradun and oversees Ayurveda, Yoga & Naturopathy, Unani, Siddha and Homeopathy. Its Director is a senior AYUSH-cadre officer, separate from the Director NHM and the Director General of Medical Health & Family Welfare. The directorate runs its own dispensaries (most of them small Ayurveda or Homeopathy units in panchayat buildings), its own colleges (notably Rishikul and Gurukul Kangri), and its own AYUSH Mission desk under the National AYUSH Mission (NAM).
What makes Uttarakhand distinctive is the depth of co-location inside NHM facilities. As reported by the directorate in 2024, AYUSH wings now operate at:
The financing arrangement at every co-located wing is structurally identical: medicines and furnishings come from the AYUSH Ministry, salaries come from NHM. This is what makes "co-location" a real programmatic instrument rather than merely a building-sharing arrangement — the AYUSH Medical Officer at a PHC is on the NHM payroll but draws Ayurveda or Homeopathy medicines from the AYUSH supply chain.
For HWC staffing, AYUSH practitioners are a major recruitment route. A BAMS, BHMS or BUMS graduate becomes eligible to serve as a Community Health Officer (the mid-level provider running every HWC) after the IGNOU Certificate in Community Health (CCH) — a six-month bridge programme covering basic clinical assessment, essential drugs, NCD screening protocols and referral pathways. The Cabinet's 2020 approval to operationalise 12,500 AYUSH-AAMs nationally rides on this same bridge cadre. In Uttarakhand, the National AYUSH Mission has approved blocks of dedicated AYUSH-AAM facilities — including 200 in Almora district alone.
Mapping the AYUSH wings against the NHM facility base yields the working share of co-location:
This is among the highest co-location rates of any state in India. For a hill state where the closest specialist may be an eight-hour drive away, having an AYUSH MO already on-site at the PHC is a meaningful access lever.
The Programme Implementation Plan is the spinal column of NHM. Every state submits a PIP each year detailing planned activities and budgets; MoHFW reviews and approves a Record of Proceedings, which becomes the operational and financial mandate for the year. Activities not in the ROP cannot be funded under NHM.
The cycle starts at the bottom. Each district prepares a District Health Action Plan (DHAP) — an activity-and-budget plan compiled by the DPMU under the CMO. Block-level inputs feed in through the BPMU. The 13 DHAPs are aggregated and reconciled at the SPMU, which compiles them into a single State PIP with cross-cutting verticals (MCH, RBSK, RKSK, NUHM, NCD, NLEP, IDSP, NMHP and so on). The PIP is signed off by the Mission Director and submitted to MoHFW's National Programme Coordination Committee (NPCC) by mid-year.
NPCC convenes a state-by-state appraisal. Each line item is examined for budgetary consistency, programmatic priority and physical-target plausibility. Iteration is normal — states are sent back for revisions and resubmissions. Once cleared, the approved version is issued as the Record of Proceedings (ROP). The first tranche of central funds (typically half of the central share) is released against the ROP and credited to the UKHFWS account.
UKHFWS releases quarterly to each of the 13 District Health Societies; DHS in turn releases monthly to Block Health Societies based on activity calendars. Block Data Managers upload monthly facility reports to HMIS by the 7th of each month; the State QA cell validates by the 15th. After the first half-year, the state files a Statement of Expenditure (SOE) and a Utilisation Certificate (UC) with MoHFW; tranche-2 release is gated on these.
A mid-year joint review is held in December or January, and supplementary PIP requests are considered if the state needs additional headroom. February and March are spent on year-end procurement, training and capital activities — bills must be cleared before 31 March, after which un-spent balances are carried forward but cannot be drawn against without re-approval. The audited Statement of Expenditure follows in the months after, and the next PIP cycle is already underway by then.
NHM is a Centrally Sponsored Scheme with a state matching share. For Uttarakhand — classified as a hill state along with the North-East — the funding ratio is 90% Centre, 10% State (against 60:40 for general states). Knowing how a rupee actually travels from MoHFW to a sub-centre helps explain why some operational levers move quickly and others don't.
The combined central-and-state pool sits at the Uttarakhand Health & Family Welfare Society — a registered society chaired by the Mission Director. UKHFWS is the legal counterparty: it holds bank accounts, signs vendor and salary contracts, and disburses to district health societies. Its annual budget is the sum of the approved PIP plus carry-forwards.
UKHFWS releases quarterly to each of the 13 District Health Societies. The DHS is chaired by the District Magistrate with the CMO as secretary. From DHS, funds flow monthly to Block Health Societies — chaired by the SDM with the BMOIC as secretary. The BHS in turn pays salaries of the Block Programme Management Unit (BPM, BAM, BDM, BCM) and contractual ANMs, and routes operational funds to facility-level Rogi Kalyan Samitis at every CHC and PHC.
Three important streams bypass the chain:
The PM-JAY hospital-insurance arm of Ayushman Bharat does not flow through UKHFWS. It runs through the State Health Authority and pays empanelled hospitals (public and private) on a case-package basis after admission. The two channels — primary-care strengthening through NHM, and secondary/tertiary insurance through PM-JAY — sit on parallel tracks even though they share the Ayushman Bharat brand.
Behind the visible architecture of facilities and people sit the quality-assurance frameworks and the vertical programmes that NHM funds. These are the rails that keep the system honest and that translate broad mandates into specific, measurable activities at the facility level.
NHM-funded facilities are assessed under three overlapping but distinct certifications:
The largest funded vertical. RCH covers JSY (cash incentive for institutional delivery), JSSK (free drugs, diagnostics, transport and diet for pregnant women and sick newborns), RBSK (school-and-Anganwadi screening for the 4 D's: defects, deficiencies, diseases, developmental delays), RKSK (adolescent health), routine immunization (now on the U-WIN platform), and the SNCU/NBSU/NRC chain for sick newborns and severely malnourished children. Front-line delivery happens at every PHC and CHC; specialised care is concentrated in the 24x7 Female Hospitals and DHs.
Hypertension, diabetes, common cancers (oral, breast, cervical), chronic respiratory disease and stroke risk are screened opportunistically at every HWC for the 30+ population. Positive cases are referred up the chain for confirmation and management. The PM National Dialysis Programme runs free dialysis units at most DHs.
Three named programmes carry the bulk: NTEP (National TB Elimination Programme — the renamed RNTCP — targeting elimination by 2025), NLEP (Leprosy), and IDSP (Integrated Disease Surveillance Programme, the all-hazards reporting system that picks up outbreak signals from every PHC). Vector-borne disease control sits on a parallel programme line.
NHM and the Ministry of Women & Child Development converge on nutrition through the POSHAN Abhiyaan. Severely malnourished children are referred from Anganwadi to Nutrition Rehabilitation Centres (NRCs) at DHs and to Malnutrition Treatment Centres at CHCs. WIFS (Weekly Iron Folic Acid Supplementation) covers school-going adolescents; ASHAs distribute IFA tablets to pregnant women and lactating mothers.
The District Mental Health Programme (DMHP) places a small mental-health team at the DH — typically a psychiatrist, a psychologist, a psychiatric social worker and a community nurse. The Tele-MANAS helpline operates as a national 24x7 backup for districts without on-site psychiatrists.
Most of this compilation describes what is funded by NHM. This section describes how much, using the Programme Implementation Plan (PIP) and Record of Proceedings (ROP) that the State submits and the Centre approves each year. Figures are bound to DATA.pipBudget in the catalog.
Since FY 2024-25, NHM Uttarakhand operates on a two-year PIP / ROP cycle. The combined allocation for FY 2024-25 and FY 2025-26 is ₹ 1,100 crore, approximately split equally across the two years. This is approved by the National Programme Coordination Committee (NPCC) at MoHFW based on the state's PIP submission.
The PIP submission breaks the resource envelope across NHM's major programme heads. Approximate shares based on aggregate of recent ROPs:
Approximate state-level annual ROP envelopes for context:
NHM is not the only source of Centre money flowing into Uttarakhand's health sector. The state also receives:
The PIP document explicitly directs districts to avoid duplication between these funding lines when scoping facility-strengthening work.
↑ Return to contentsAll facts in this document are traceable to official government sources, audit reports, and the NHM Uttarakhand directory. References are listed in order of first reference within the document.
Every acronym, programme name, and Hindi/Sanskrit term used in this compilation, with its full expansion and a one-line definition.
Every figure cited in this compilation is traceable to an official, audit, or peer-reviewed source. Every visual element on the page — map tiles, fonts, charts, polygons — comes from open or open-licensed software. This page is the audit trail for both halves.
Every numeric claim in the dashboard, reference text and glossary should be reproducible. The recipe:
If a figure on this page does not match the source, the source is correct and this page is wrong. The thumbs-down feedback button at the bottom right of the chat sends corrections directly to the build chain.
Two halves to this tab. First, the live mechanism — one click triggers the refresh pipeline and you can watch counters across the dashboard update in place. Then, the journey behind it: fifteen phases of conversation between a human stakeholder and an AI build agent, with the real prompts that shaped each phase, the architectural decisions that emerged, and the patterns that fell out. Useful as a learning artifact if you want to build a similar self-maintaining dashboard. For the technology stack itself — FOSS pipeline, mapping tools, typography, build philosophy — see the Tech tab.
The DATA catalog at the top of the script can be refreshed manually (Section F below). It can also, in principle, be refreshed by an automated agent — the same kind of agent that built this site. The button below simulates that pipeline end-to-end and actually applies the resulting diff to the in-memory window.DATA object — scroll up to the Overview tab after running it to watch the affected counters refresh in place. The network calls remain simulated (a static HTML file cannot reach MoHFW); the diff is a fixed illustrative payload, applied for demonstration. The change is in-memory only — a page reload restores the catalog baseline, or click Reset to catalog to revert without reloading.
This site was not authored top-to-bottom in one pass. It was constructed in a long iterative dialogue between a human stakeholder and an AI build agent — the same kind of agent capable of running tools (web search, code execution, file operations) inside a single conversation context. The agent's loop, tool use, and the human-in-the-loop refinement pattern are diagrammed below.
Each phase moved the artifact forward by one bounded request. The agent never had to guess what was next, and the artifact never had to be rebuilt — it grew incrementally.
Block diagram of the loop. Run it once and you get a small, well-bounded change. Run it 15 times and you get the artifact you're reading. The discipline is in keeping each iteration small enough that the agent can't miss what was asked.
Looking back at fifteen phases, the prompts that moved this artifact forward fastest had common shapes. Six recur often enough to be worth naming.
In the spirit of honest documentation: three patterns that produced less efficient sessions.
Five layers, each doing one thing. A number in the JSON catalog flows up to a counter on the screen. A click on the pipeline trigger flows down to mutate the catalog. The same architecture handles both manual edits and agent-driven refreshes.
window.DATA at first render, mounts HTML into div placeholders by ID, attaches its own event handlers[data-bind] element in the DOM, resolves the dot-path against window.DATA, formats per data-bind-format, replaces text content. Runs at page load and after every diff applydistricts, blocksByDistrict) into module scope, exposes the rest as window.DATA#pipelineBtn in the Build tabapplyDiff() walks SIM_DIFF and writes each value into window.DATA via dot-pathapplyDataBindings() fires — every [data-bind] element in the DOM gets re-read.value-updated class — CSS keyframes pulse the cell greenA real agent run replaces step 3's hardcoded SIM_DIFF with values fetched from MoHFW / AYUSH / NHM endpoints. Everything else is identical. The pipeline architecture doesn't change between "simulation" and "production" — only the source of the diff.
Six patterns that fell out of the journey. Generalizable to any reference dashboard that needs to stay accurate over time.
<span data-bind="path.to.value">fallback</span> in HTML. Have one renderer walk the DOM and replace text. Don't sprinkle document.querySelector(...).textContent = ... across the codebase. The data-bind system is essentially a one-direction Vue-style binding in 40 lines of vanilla JS.data-source attributes at render time, so every number on the page can show "where did this come from" on hover or in print. Costs nothing; pays off the first time someone asks "is this current?"window.DATA at runtime. Same data structure, two entry points. The architecture treats them as identical.window.DATA only — they don't get persisted. Reload restores the baseline. A "Reset to catalog" button does the same in-session. This means simulated or experimental updates can never poison the document. Cache truth (Build button stats from live DOM), don't cache fiction (pipeline sim diffs).querySelectorAll. The reported stats can never lie about themselves because they're computed from the live structure. The catalog holds a baseline; the Build button confirms reality. Useful for any artifact where "is the metadata accurate?" matters.The combined effect of P1–P6: this file edits like a spreadsheet, refreshes like a CMS, and ships like a static file. No build step, no server, no API key. Open the HTML in any text editor; the data is at the top; everything propagates on next reload.
Every editable number on the page is wired to a single DATA catalog at the top of the script block in this HTML file. To refresh figures (e.g. when MoHFW publishes new AAM data, or the ASHA workforce grows), the workflow is short:
DATA CATALOG — you'll land on the documented data block at the top of the script.state, workforce, aam, ayush, norms, yqwp. Each property has a one-line comment explaining what it represents.12018, not 12,018 — the renderer adds commas automatically).data-bind="path.to.value" attribute. On page load, a small renderer walks the DOM and replaces text content from the DATA object. The HTML still ships with fallback values, so the page renders correctly even if the script fails.<span data-bind="newPath">fallback</span>. Optional: data-bind-format="pct" for percentages, data-bind-format="plain" for un-formatted integers.Two layers of technology shipped with this artifact. What's used today: a small set of mapping, charting, typography and CSS choices that produced the file you're reading. What it could become: a full open-source agentic pipeline that re-fetches government data, validates it, regenerates the HTML, and deploys the site — on infrastructure you own, with zero SaaS dependencies and no commercial-LLM API keys leaving your network. Section A documents that future-state pipeline; Section B catalogues the tools used to ship the current artifact.
An autonomous, scheduled pipeline of free and open-source software can re-fetch the underlying government data, validate it, generate the HTML, and deploy the site — on infrastructure you own, with no SaaS dependencies and no commercial-LLM API keys leaving your network. The architecture below shows how to build that agent end-to-end. Sections D and E later in this tab show the same workflow as a live simulation and as a manual recipe respectively.
Each role in the pipeline has multiple FOSS options — pick the one that matches your team’s preferences and infrastructure constraints.
Visual node-based workflow builder. Native nodes for HTTP, Postgres, S3, OpenAI/Ollama, Slack, cron triggers. A 700+ node ecosystem covers most government portals out of the box. Run as a single Docker container.
Drop-in alternative if you want a fully OSI-approved licence. Same node-based UX, smaller ecosystem but covers the essentials (HTTP, Postgres, OpenAI-compatible LLM, webhook).
If your team is Python-comfortable and prefers code-as-config. DAGs in Python, web UI, deep observability hooks, mature scheduler. Heavier to operate than n8n.
For pipelines that need crash-recovery semantics — if a stage 5 LLM call dies mid-flight, Temporal resumes from the last completed activity. Overkill for weekly refreshes; essential for high-volume pipelines.
Explicit state machines for agentic flows. Each node is a callable that mutates state; edges define routing. Best fit for the reconciliation + critic stages where flow control matters more than free-form chat. Works with any OpenAI-compatible endpoint.
Model the pipeline as a crew of specialised agents (Researcher, Validator, Writer, Critic) rather than a graph. Higher abstraction; less control. Good for prototyping the agentic shape before committing to LangGraph.
Stronger on RAG and document ingestion than agentic flow. Useful in stage 4 (extraction): point LlamaIndex at the raw artifact bucket, get back structured documents indexed with metadata.
Cleaner abstractions than LangChain, smaller community. A good fit if you prefer typed pipelines over implicit chains.
Simplest path to running Llama 3.3, Mistral, Qwen, Gemma 3 or DeepSeek locally. Single binary, OpenAI-compatible API on port 11434. Auto-detects Apple Silicon, NVIDIA and AMD GPUs. Smallest possible operational footprint.
PagedAttention memory management lets a single A100 / H100 serve dozens of concurrent requests. Production choice when latency and throughput matter; needs more setup than Ollama.
The engine under Ollama. Use directly if you want fine control over quantization (Q4_K_M, Q5_K_S, etc.), CPU-only deployments, or edge devices.
Single API endpoint that wraps multiple backend engines (llama.cpp, transformers, diffusers). Useful when you want to swap models without changing client code.
Strong instruction-following, ~70 GB in 4-bit quant (single A100 or 2× RTX 3090). Best baseline for the LLM critic and extraction-fallback roles. The 8B variant fits on a 16 GB GPU and is plenty for structured extraction.
Mistral 7B and Mixtral 8x7B / 8x22B are released under permissive Apache 2.0 — no commercial-use ambiguity. Mixtral’s mixture-of-experts gives strong quality at moderate VRAM.
Excellent at JSON-mode generation and tool calling. The 32B variant is a sweet spot for structured-extraction agents. Trained with strong multilingual data — useful for Indic-language source documents.
For pipelines that need deep reasoning (e.g. the critic agent comparing multi-source contradictions). R1 is reasoning-tuned. Hardware demand is real — quantize or rent compute and self-host the wrapper.
Headless Chromium, Firefox or WebKit driven from Python or Node. Required for any portal that uses client-side JS to render tables (most newer ones do). Auto-wait, network interception, screenshots for visual diff testing.
Wraps Playwright/Cheerio with crawl queues, retry logic, request deduplication, and proxy rotation. The right abstraction when you go from one-off scripts to "crawl 200 pages reliably every week".
If your sources are mostly static HTML and you want maximum throughput, Scrapy beats Playwright by an order of magnitude. Spiders are declarative; pipelines compose.
Marker (Datalab) and Docling (IBM) both turn PDFs into clean Markdown with tables, footnotes, and headings preserved. Critical for the Rural Health Statistics PDF pipeline.
Holds the versioned DATA catalog (one row per metric per run-id), the audit log, and — with pgvector — semantic embeddings of every source document for retrieval-augmented critic checks.
Single binary that gives you a production-ready S3 endpoint. Stores raw HTML, PDFs and processing artifacts under run-id keys. AGPL means commercial users may want to weigh alternatives.
Garage (Deuxfleurs) and SeaweedFS are smaller-footprint S3 stores. SeaweedFS handles billions of small files; Garage focuses on geo-distributed setups. Pick if MinIO’s licence concerns you.
For semantic search over historical source documents. The critic agent uses Qdrant to retrieve "what did the AAM dashboard say last quarter?" and compare. Single Docker image, fast filtered search.
Pa11y runs WCAG checks against the rendered page; html5lib parses and validates structure. Both run headless in CI; both block deploy on regressions.
Logs every prompt, completion, latency, token count and tool call across the LangGraph runs. Self-hostable Postgres-backed, OpenTelemetry-compatible. The right answer if “why did the critic flag run #142?” needs to be debuggable months later.
One-line config, automatic Let’s Encrypt, HTTP/3 by default. Serves the built index.html over HTTPS without you ever touching certificate renewal.
Self-hosted push notification service. The critic agent posts to a topic when human review is required. Clients on phone, desktop, browser. Replaces Slack webhook for teams that want nothing on commercial infrastructure.
Versioned storage for the generated index.html. Forgejo is the community fork of Gitea. Either runs as a single Go binary alongside the rest of the stack.
Prometheus scrapes pipeline metrics. Grafana renders dashboards. Loki stores logs. Standard CNCF stack, runs entirely on your infrastructure, OpenTelemetry-compatible.
A simplified version of what the n8n canvas looks like once the pipeline is wired. Every node has a green status dot when it last ran successfully; a red dot if a human is required.
A single docker-compose.yml stands up the entire stack on a 4-core / 16 GB box (8-core / 32 GB if hosting LLMs locally). Each service is independently restartable.
# docker-compose.yml — FOSS agentic build stack
services:
n8n: # workflow orchestrator
image: n8nio/n8n:latest
environment: { DB_TYPE: postgresdb, DB_POSTGRESDB_HOST: postgres }
ports: ['5678:5678']
postgres: # canonical DATA store + n8n state
image: postgres:17
volumes: ['./pg:/var/lib/postgresql/data']
minio: # raw artifact storage (S3-compatible)
image: minio/minio:latest
command: server /data --console-address ':9001'
ports: ['9000:9000', '9001:9001']
ollama: # local LLM inference
image: ollama/ollama:latest
volumes: ['./ollama:/root/.ollama']
qdrant: # vector store for the critic
image: qdrant/qdrant:latest
langfuse: # LLM observability
image: langfuse/langfuse:latest
caddy: # serves the built index.html
image: caddy:latest
volumes: ['./site:/srv', './Caddyfile:/etc/caddy/Caddyfile']
ports: ['80:80', '443:443']
prometheus: # metrics
image: prom/prometheus:latest
grafana: # dashboards
image: grafana/grafana-oss:latest
ports: ['3000:3000']
JavaScript library that renders the interactive choropleth, district markers, permanent labels, and the hover/click polygon interactions in Exhibit A. Loaded from jsDelivr CDN.
The base-map tile layer behind the choropleth — rendered at 0.55 opacity so the warm-tone district fills dominate while terrain and labels remain visible underneath.
Stacked horizontal bar chart in Programs · Exhibit A (AAM Rollout). Loaded from jsDelivr CDN. The Facility Distribution panel (C) replaced its Chart.js bar with a hand-built sortable HTML table.
The three Plex sub-families do all the heavy lifting: Serif for headings and reading prose, Sans for body and UI text, Mono for KPIs, labels, hash codes and rank numbers. Loaded from Google Fonts as a single family request.
Origin for Leaflet, Chart.js, and the Uttarakhand district GeoJSON. Pinning specific commits in URLs guards against silent upstream changes.
Serves the Plex family with browser-cached subsets. Single stylesheet request brings in three weights across three sub-families.
All HTML, CSS and JavaScript live in one document. No React, no Vue, no bundler, no transpilation step. The only external runtime requests are the three CDN assets (Leaflet, Chart.js, Google Fonts) and the GeoJSON. Opens in any modern browser, can be saved offline, can be archived as a single artifact.