Analysis, Evaluation, Rebuild

Bahmni and Legacy Code Migration

Bahmni is an open-source Hospital Administration
System & EMR constructed to assist healthcare supply in low-resource
settings offering a wealthy interface for scientific and administrative customers.
The Bahmni frontend was initially
developed utilizing AngularJS (model 1.x)—an
early however highly effective framework for constructing dynamic net purposes.
Nevertheless, AngularJS has lengthy been deprecated by the Angular crew at Google,
with official long-term assist having led to December 2021.

Regardless of this, Bahmni continues to rely closely on AngularJS for a lot of of
its core workflows. This reliance introduces important dangers, together with
safety vulnerabilities from unpatched dependencies, problem in
onboarding builders unfamiliar with the outdated framework, restricted
compatibility with fashionable instruments and libraries, and diminished maintainability
as new necessities are constructed on an growing older codebase.

In healthcare techniques, the continued use of outdated software program can
adversely have an effect on scientific workflows and compromise affected person knowledge security.
For Bahmni, frontend migration has develop into a essential precedence.

Analysis, Evaluation, Rebuild

The workflow I adopted known as “Analysis, Evaluation, Rebuild” — the place
we do a characteristic migration analysis utilizing a few MCP servers, validate
and approve the strategy AI proposes, rebuild the characteristic after which as soon as
all of the code era is finished, refactor issues that you just did not like.

The Workflow

1. Put together a listing of options focused for migration. Choose one characteristic to
   start with.
2. Use Mannequin Context Protocol (MCP) servers to analysis the chosen characteristic
   by producing a contextual evaluation of the chosen characteristic by way of a Massive
   Language Mannequin (LLM).
3. Have area specialists assessment the generated evaluation, making certain it’s
   correct, aligns with present challenge conventions and architectural tips.
   If the characteristic isn’t sufficiently remoted for migration, defer it and replace
   the characteristic listing accordingly.
4. Proceed with LLM-assisted rebuild of the validated characteristic to the goal
   system or framework.
5. Till the listing is empty, return to #2

Earlier than Getting Began

Earlier than we proceed with the workflow, it’s important to have a
high-level understanding of the present codebase and decide which
elements needs to be retained, discarded, or deferred for future
consideration.

Within the context of Bahmni, Show
Controls
are modular, configurable widgets that may be embedded throughout numerous
pages to reinforce the system’s flexibility. Their decoupled nature makes
them well-suited for focused modernization efforts. Bahmni at the moment
consists of over 30 show controls developed over time. These controls are
extremely configurable, permitting healthcare suppliers to tailor the interface
to show pertinent knowledge like diagnoses, remedies, lab outcomes, and
extra. By leveraging show controls, Bahmni facilitates a customizable
and streamlined consumer expertise, aligning with the varied wants of
healthcare settings.

All the present Bahmni show controls are constructed over OpenMRS REST
endpoint, which is tightly coupled with the OpenMRS knowledge mannequin and
particular implementation logic. OpenMRS (Open
Medical Document System) is an open-source platform designed to function a
foundational EMR system primarily for low-resource environments offering
customizable and scalable methods to handle well being knowledge, particularly in
creating nations. Bahmni is constructed on high of OpenMRS, counting on
OpenMRS for scientific knowledge modeling and affected person file administration, utilizing
its APIs and knowledge constructions. When somebody makes use of Bahmni, they’re
basically utilizing OpenMRS as half of a bigger system.

FHIR (Quick Healthcare
Interoperability Assets) is a contemporary customary for healthcare knowledge
alternate, designed to simplify interoperability by utilizing a versatile,
modular strategy to symbolize and share scientific, administrative, and
monetary knowledge throughout techniques. FHIR was launched by
HL7 (Well being Stage Seven Worldwide), a
not-for-profit requirements improvement group that performs a pivotal
position within the healthcare business by creating frameworks and requirements for
the alternate, integration, sharing, and retrieval of digital well being
data. The time period “Well being Stage Seven” refers back to the seventh layer
of the OSI (Open Techniques
Interconnection) mannequin—the applying
layer,
answerable for managing knowledge alternate between distributed techniques.

Though FHIR was initiated in 2011, it reached a big milestone
in December 2018 with the discharge of FHIR Launch 4 (R4). This launch
launched the primary normative content material, marking FHIR’s evolution right into a
secure, production-ready customary appropriate for widespread adoption.

Bahmni’s improvement commenced in early 2013, throughout a time when FHIR
was nonetheless in its early levels and had not but achieved normative standing.
As such, Bahmni relied closely on the mature and production-proven OpenMRS
REST API. Given Bahmni’s dependence on OpenMRS, the supply of FHIR
assist in Bahmni was inherently tied to OpenMRS’s adoption of FHIR. Till
lately, FHIR assist in OpenMRS remained restricted, experimental, and
lacked complete protection for a lot of important useful resource sorts.

With the latest developments in FHIR assist inside OpenMRS, a key
precedence within the ongoing migration effort is to architect the goal system
utilizing FHIR R4. Leveraging FHIR endpoints facilitates standardization,
enhances interoperability, and simplifies integration with exterior
techniques, aligning the system with globally acknowledged healthcare knowledge
alternate requirements.

For the aim of this experiment, we are going to focus particularly on the
Therapies Show Management as a consultant candidate for
migration.

Determine 2: Legacy Therapies Show Management constructed utilizing
Angular and built-in with OpenMRS REST endpoints

The Remedy Particulars Management is a selected sort of show management
in Bahmni that focuses on presenting complete details about a
affected person’s prescriptions or drug orders over a configurable variety of
visits. This management is instrumental in offering clinicians with a
consolidated view of a affected person’s remedy historical past, aiding in knowledgeable
decision-making. It retrieves knowledge by way of a REST API, processing it right into a
view mannequin for UI rendering in a tabular format, supporting each present
and historic remedies. The management incorporates error dealing with, empty
state administration, and efficiency optimizations to make sure a sturdy and
environment friendly consumer expertise.

The info for this management is sourced from the
/openmrs/ws/relaxation/v1/bahmnicore/drugOrders/prescribedAndActive endpoint,
which returns visitDrugOrders. The visitDrugOrders array accommodates
detailed entries that hyperlink drug orders to particular visits, together with
metadata concerning the supplier, drug idea, and dosing directions. Every
drug order consists of prescription particulars reminiscent of drug identify, dosage,
frequency, period, administration route, begin and cease dates, and
customary code mappings (e.g., WHOATC, CIEL, SNOMED-CT, RxNORM).

Here’s a pattern JSON response from Bahmni’s
/bahmnicore/drugOrders/prescribedAndActive REST API endpoint containing
detailed details about a affected person’s drug orders throughout a selected
go to, together with metadata like drug identify, dosage, frequency, period,
route, and prescribing supplier.

{
  "visitDrugOrders": [
    {
      "visit": {
        "uuid": "3145cef3-abfa-4287-889d-c61154428429",
        "startDateTime": 1750033721000
      },
      "drugOrder": {
        "concept": {
          "uuid": "70116AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA",
          "name": "Acetaminophen",
          "dataType": "N/A",
          "shortName": "Acetaminophen",
          "units": null,
          "conceptClass": "Drug",
          "hiNormal": null,
          "lowNormal": null,
          "set": false,
          "mappings": [
            {
              "code": "70116",
              "name": null,
              "source": "CIEL"
            },y
            /* Response Truncated */
          ]
        },
        "directions": null,
        "uuid": "a8a2e7d6-50cf-4e3e-8693-98ff212eee1b",

        "orderType": "Drug Order",
        "accessionNumber": null,
        "orderGroup": null,
        "dateCreated": null,
        "dateChanged": null,
        "dateStopped": null,
        "orderNumber": "ORD-1",
        "careSetting": "OUTPATIENT",
        "motion": "NEW",
        "commentToFulfiller": null,
        "autoExpireDate": 1750206569000,
        "urgency": null,
        "previousOrderUuid": null,
        "drug": {
          "identify": "Paracetamol 500 mg",
          "uuid": "e8265115-66d3-459c-852e-b9963b2e38eb",
          "type": "Pill",
          "power": "500 mg"
        },
        "drugNonCoded": null,
        "dosingInstructionType": "org.openmrs.module.bahmniemrapi.drugorder.dosinginstructions.FlexibleDosingInstructions",
        "dosingInstructions": {
          "dose": 1.0,
          "doseUnits": "Pill",
          "route": "Oral",
          "frequency": "Twice a day",
          "asNeeded": false,
          "administrationInstructions": "{"directions":"As directed"}",
          "amount": 4.0,
          "quantityUnits": "Pill",
          "numberOfRefills": null
        },
        "dateActivated": 1750033770000,
        "scheduledDate": 1750033770000,
        "effectiveStartDate": 1750033770000,
        "effectiveStopDate": 1750206569000,
        "orderReasonText": null,
        "period": 2,
        "durationUnits": "Days",
        "voided": false,
        "voidReason": null,
        "orderReasonConcept": null,
        "sortWeight": null,
        "conceptUuid": "70116AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA"
      },
      "supplier": {
        "uuid": "d7a67c17-5e07-11ef-8f7c-0242ac120002",
        "identify": "Tremendous Man",
        "encounterRoleUuid": null
      },
      "orderAttributes": null,
      "retired": false,
      "encounterUuid": "fe91544a-4b6b-4bb0-88de-2f9669f86a25",
      "creatorName": "Tremendous Man",
      "orderReasonConcept": null,
      "orderReasonText": null,
      "dosingInstructionType": "org.openmrs.module.bahmniemrapi.drugorder.dosinginstructions.FlexibleDosingInstructions",
      "previousOrderUuid": null,
      "idea": {
        "uuid": "70116AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA",
        "identify": "Acetaminophen",
        "dataType": "N/A",
        "shortName": "Acetaminophen",
        "models": null,
        "conceptClass": "Drug",
        "hiNormal": null,
        "lowNormal": null,
        "set": false,
        "mappings": [
          {
            "code": "70116",
            "name": null,
            "source": "CIEL"
          },
          /* Response Truncated */
        ]
      },
      "sortWeight": null,
      "uuid": "a8a2e7d6-50cf-4e3e-8693-98ff212eee1b",
      "effectiveStartDate": 1750033770000,
      "effectiveStopDate": 1750206569000,
      "orderGroup": null,
      "autoExpireDate": 1750206569000,
      "scheduledDate": 1750033770000,
      "dateStopped": null,
      "directions": null,
      "dateActivated": 1750033770000,
      "commentToFulfiller": null,
      "orderNumber": "ORD-1",
      "careSetting": "OUTPATIENT",
      "orderType": "Drug Order",
      "drug": {
        "identify": "Paracetamol 500 mg",
        "uuid": "e8265115-66d3-459c-852e-b9963b2e38eb",
        "type": "Pill",
        "power": "500 mg"
      },
      "dosingInstructions": {
        "dose": 1.0,
        "doseUnits": "Pill",
        "route": "Oral",
        "frequency": "Twice a day",
        "asNeeded": false,
        "administrationInstructions": "{"directions":"As directed"}",
        "amount": 4.0,
        "quantityUnits": "Pill",
        "numberOfRefills": null
      },
      "durationUnits": "Days",
      "drugNonCoded": null,
      "motion": "NEW",
      "period": 2
    }
  ]
}

The /bahmnicore/drugOrders/prescribedAndActive mannequin differs considerably
from the OpenMRS FHIR
MedicationRequest
mannequin in each construction and illustration. Whereas the Bahmni REST mannequin is
tailor-made for UI rendering with visit-context grouping and consists of
OpenMRS-specific constructs like idea, drug, orderNumber, and versatile
dosing directions, the FHIR MedicationRequest mannequin adheres to worldwide
requirements with a normalized, reference-based construction utilizing assets reminiscent of
Remedy, Encounter, Practitioner, and coded parts in
CodeableConcept and Timing.

Analysis

The “Analysis” part of the strategy includes producing an
MCP-augmented LLM evaluation of the chosen Show Management. This part is
centered round understanding the legacy system’s conduct by inspecting
its supply code and conducting reverse engineering. Such evaluation is
important for informing the ahead engineering efforts. Whereas not all
recognized necessities could also be carried ahead—notably in long-lived
techniques the place sure functionalities could have develop into out of date—it’s
essential to have a transparent understanding of present behaviors. This allows
groups to make knowledgeable choices about which parts to retain, discard,
or redesign within the goal system, making certain that the modernization effort
aligns with present enterprise wants and technical objectives.

At this stage, it’s useful to take a step again and think about how human
builders sometimes strategy a migration of this nature. One key perception
is that migrating from Angular to React depends closely on contextual
understanding. Builders should draw upon numerous dimensions of information
to make sure a profitable and significant transition. The essential areas of
focus sometimes embody:

• Objective Analysis: understanding the useful intent and position of the
  present Angular elements throughout the broader software.
• Information Mannequin Evaluation: reviewing the underlying knowledge constructions and their
  relationships to evaluate compatibility with the brand new structure.
• Information Circulation Mapping: tracing how knowledge strikes from backend APIs to the
  frontend UI to make sure continuity within the consumer expertise.
• FHIR Mannequin Alignment: figuring out whether or not the present knowledge mannequin will be
  mapped to an HL7 FHIR-compatible construction, the place relevant.
• Comparative Evaluation: evaluating structural and useful similarities,
  variations, and potential gaps between the previous and goal implementations.
• Efficiency Concerns: making an allowance for areas for efficiency
  enhancement within the new system.
• Characteristic Relevance: assessing which options needs to be carried ahead,
  redesigned, or deprecated based mostly on present enterprise wants.

This context-driven evaluation is commonly probably the most difficult side of
any legacy migration. Importantly, modernization isn’t merely about
changing outdated applied sciences—it’s about reimagining the way forward for the
system and the enterprise it helps. It includes the evolution of the
software throughout its complete lifecycle, together with its structure, knowledge
constructions, and consumer expertise.

The experience of material specialists (SMEs) and area specialists
is essential to grasp present conduct and to arrange a information for the
migration. And what higher approach to seize the anticipated conduct than
by way of well-defined check situations towards which the migrated code will
be evaluated. Understanding what situations are to be examined is essential
not simply in ensuring that – the whole lot that used to work nonetheless works
and the brand new conduct would work as anticipated but additionally as a result of now your LLM
has a clearly outlined set of objectives that it is aware of is what’s anticipated. By
defining these objectives explicitly, we will make the LLM’s responses as
deterministic as doable, avoiding the unpredictability of probabilistic
responses and making certain extra dependable outcomes through the migration
course of.

Based mostly on this understanding, I developed a complete and
strategically structured immediate
designed to seize all related data successfully.

Whereas the immediate covers all anticipated areas—reminiscent of knowledge stream,
configuration, key features, and integration—it additionally consists of a number of
sections that warrant particular point out:

• FHIR Compatibility: this part maps the customized Bahmni knowledge mannequin
  to HL7 FHIR assets and highlights gaps, thereby supporting future
  interoperability efforts. Finishing this mapping requires a stable understanding
  of FHIR ideas and useful resource constructions, and could be a time-consuming job. It
  sometimes includes a number of hours of detailed evaluation to make sure correct
  alignment, compatibility verification, and identification of divergences between
  the OpenMRS and FHIR medicine fashions, which may now be finished in a matter of
  seconds.
• Testing Tips for React + TypeScript Implementation Over OpenMRS
  FHIR: this part gives structured check situations that emphasize knowledge
  dealing with, rendering accuracy, and FHIR compliance for the modernized frontend
  elements. It serves as a superb basis for the event course of,
  setting out a compulsory set of standards that the LLM ought to fulfill whereas
  rebuilding the element.
• Customization Choices: this outlines accessible extension factors and
  configuration mechanisms that improve maintainability and flexibility throughout
  various implementation situations. Whereas a few of these choices are documented,
  the LLM-generated evaluation usually uncovers extra customization paths
  embedded within the codebase. This helps establish legacy customization approaches
  extra successfully and ensures a extra exhaustive understanding of present
  capabilities.

To collect the mandatory knowledge, I utilized two light-weight servers:

• An Atlassian MCP server to extract any accessible documentation on the
  show management.
• A filesystem MCP server, the place the legacy frontend code and configuration
  have been mounted, to supply supply code-level evaluation.

Determine 3: MCP + Cline + Claude Setup Diagram

Whereas elective, this filesystem server allowed me to give attention to the goal
system’s code inside my IDE, with the legacy reference codebases conveniently
accessible by way of the mounted server.

These mild weight servers every expose particular capabilities by way of the
standardized Mannequin Context Protocol, which is then utilized by Cline (my shopper in
this case) to entry the code base, documentation and configuration. Because the
configurations shipped are opinionated and the paperwork usually outdated, I added
particular directions to take the supply code as the one supply of reality and
the remainder as a supplementary reference.

Evaluation

The second part of the strategy —is the place the human within the loop
turns into invaluable.

The AI-generated evaluation is not meant to be accepted at face worth,
particularly for complicated codebases. You’ll nonetheless want a website skilled and an
architect to vet, contextualize, and information the migration course of. AI alone
is not going emigrate a complete challenge seamlessly; it requires
considerate decomposition, clear boundaries, and iterative validation.

Not all these necessities will essentially be integrated into the
goal system, for instance the power to print a prescription sheet based mostly
on the drugs prescribed is deferred for now.

On this case, I augmented the evaluation with pattern responses from the
FHIR endpoint and whereas discarding points of the system that aren’t
related to the modernization effort. This consists of efficiency
optimizations, check circumstances that aren’t immediately related to the migration,
and configuration choices such because the variety of rows to show and
whether or not to point out energetic or inactive drugs. I felt these will be
addressed as a part of the subsequent iteration.

As an illustration, think about the unit check situations outlined for rendering
remedy knowledge:

        ✅ Comfortable Path

        It ought to accurately render the drugName column.
        It ought to accurately render the standing column with the suitable Tag colour.
        It ought to accurately render the precedence column with the proper precedence Tag.
        It ought to accurately render the supplier column.
        It ought to accurately render the startDate column.
        It ought to accurately render the period column.
        It ought to accurately render the frequency column.
        It ought to accurately render the route column.
        It ought to accurately render the doseQuantity column.
        It ought to accurately render the instruction column.

        ❌ Unhappy Path

        It ought to present a “-” if startDate is lacking.
        It ought to present a “-” if frequency is lacking.
        It ought to present a “-” if route is lacking.
        It ought to present a “-” if doseQuantity is lacking.
        It ought to present a “-” if instruction is lacking.
        It ought to deal with circumstances the place the row knowledge is undefined or null.
      

Changing lacking values with “-” within the unhappy path situations has been eliminated,
because it doesn’t align with the necessities of the goal system. Such choices
needs to be guided by enter from the subject material specialists (SMEs) and
stakeholders, making certain that solely performance related to the present enterprise
context is retained.

The literature gathered on the show management now must be coupled with
challenge conventions, practices, and tips with out which the LLM is open to
interpret the above request, on the information that it was skilled with. This consists of
entry to features that may be reused, pattern knowledge fashions and companies and
reusable atomic elements that the LLMs can now depend on. If such practices,
model guides and tips aren’t clearly outlined, each iteration of the
migration dangers producing non-conforming code. Over time, this could contribute to
a fragmented codebase and an accumulation of technical debt.

The core goal is to outline clear, project-specific coding requirements and
model guides to make sure consistency within the generated code. These requirements act as
a foundational reference for the LLM, enabling it to supply output that aligns
with established conventions. For instance, the Google TypeScript Fashion Information can
be summarized and documented as a TypeScript model information saved within the goal
codebase. This file is then learn by Cline in the beginning of every session to make sure
that every one generated TypeScript code adheres to a constant and acknowledged
customary.

Rebuild

Rebuilding the characteristic for a goal system with LLM-generated code is
the ultimate part of the workflow. Now with all of the required knowledge gathered,
we will get began with a easy immediate

You’re tasked with constructing a Remedy show management within the new react ts fhir frontend. You could find the main points of the legacy Remedy show management implementation in docs/treatments-legacy-implementation.md. Create the brand new show management by following the docs/display-control-guide.md

At this stage, the LLM generates the preliminary code and check situations,
leveraging the knowledge supplied. As soon as this output is produced, it’s
important for area specialists and builders to conduct a radical code assessment
and apply any needed refactoring to make sure alignment with challenge requirements,
performance necessities, and long-term maintainability.

Refactoring the LLM-generated code is essential to making sure the code stays
clear and maintainable. With out correct refactoring, the end result could possibly be a
disorganized assortment of code fragments fairly than a cohesive, environment friendly
system. Given the probabilistic nature of LLMs and the potential discrepancies
between the generated code and the unique aims, it’s important to
contain area specialists and SMEs at this stage. Their position is to totally
assessment the code, validate that the output aligns with the preliminary expectations,
and assess whether or not the migration has been efficiently executed. This skilled
involvement is essential to make sure the standard, accuracy, and general success of
the migration course of.

This part needs to be approached as a complete code assessment—just like
reviewing the work of a senior developer who possesses sturdy language and
framework experience however lacks familiarity with the particular challenge context.
Whereas technical proficiency is crucial, constructing sturdy techniques requires a
deeper understanding of domain-specific nuances, architectural choices, and
long-term maintainability. On this context, the human-in-the-loop performs a
pivotal position, bringing the contextual consciousness and system-level understanding
that automated instruments or LLMs could lack. It’s a essential course of to make sure that
the generated code integrates seamlessly with the broader system structure
and aligns with project-specific necessities.

In our case, the intent and context of the rebuild have been clearly outlined,
which minimized the necessity for post-review refactoring. The necessities gathered
through the analysis part—mixed with clearly articulated challenge conventions,
expertise stack, coding requirements, and magnificence guides—ensured that the LLM had
minimal ambiguity when producing code. Because of this, there was little left for
the LLM to deduce independently.

That stated, any unresolved questions concerning the implementation plan can
result in deviations from the anticipated output. Whereas it isn’t possible to
anticipate and reply each such query upfront, it is very important
acknowledge the inevitability of “unknown unknowns.” That is exactly the place a
thorough assessment turns into important.

On this specific occasion, my familiarity with the show management we have been
rebuilding allowed me to proactively decrease such unknowns. Nevertheless, this degree
of context could not all the time be accessible. Subsequently, I strongly suggest
conducting an in depth code assessment to assist uncover these hidden gaps. If
recurring points are recognized, the immediate can then be refined to handle them
preemptively in future iterations.

The attract of LLMs is plain; they provide a seemingly easy answer
to complicated issues, and builders can usually create such an answer shortly and
without having years of deep coding expertise. This could not create a bias
within the specialists, succumbing to the attract of LLMs and finally take their arms
off the wheel.

Consequence

Determine 4: A excessive degree overview of the method; taking a characteristic from the legacy codebase and utilizing LLM-assisted evaluation to rebuild it throughout the goal system

In my case the code era course of took about 10 minutes to
full. The evaluation and implementation, together with each unit and
integration assessments with roughly 95% protection, have been accomplished utilizing
Claude 3.5 Sonnet (20241022). The full value for this effort was about
$2.

Determine 5: Legacy Therapies Show Management constructed utilizing Angular and built-in with OpenMRS REST endpoints

Determine 6: Modernized Therapies Show Management rebuilt
utilizing React and TypeScript, leveraging FHIR endpoints

With out AI assist, each the technical evaluation and implementation
would have doubtless taken a developer a minimal of two to 3 days. In my
case, creating a reusable, general-purpose immediate—grounded within the shared
architectural ideas behind the roughly 30 show controls in
Bahmni—took about 5 targeted iterations over 4 hours, at a barely
greater inference value of round $10 throughout these cycles. This effort was
important to make sure the generated immediate was modular and broadly
relevant, given that every show management in Bahmni is basically a
configurable, embeddable widget designed to reinforce system flexibility
throughout completely different scientific dashboards.

Even with AI-assisted era, one of many key prices in improvement
stays the time and cognitive load required to investigate, assessment, and
validate the output. Due to my prior expertise with Bahmni, I used to be in a position
to assessment the generated evaluation in below quarter-hour, supplementing it
with fast parallel analysis to validate the claims and knowledge mappings. I
was pleasantly shocked by the standard of the evaluation: the information mannequin
mapping was exact, the logic for transformation was sound, and the check
case strategies coated a complete vary of situations, each typical
and edge circumstances.

Code assessment, nonetheless, emerged as probably the most important problem.
Reviewing the generated code line by line throughout all modifications took me
roughly 20 minutes. In contrast to pairing with a human developer—the place
iterative discussions happen at a manageable tempo—working with an AI system
able to producing complete modules inside seconds creates a bottleneck
on the human aspect, particularly when making an attempt line-by-line scrutiny. This
isn’t a limitation of the AI itself, however fairly a mirrored image of human
assessment capability. Whereas AI-assisted code reviewers are sometimes proposed as a
answer, they’ll usually establish syntactic points, adherence to finest
practices, and potential anti-patterns—however they battle to evaluate intent,
which is essential in legacy migration initiatives. This intent, grounded in
area context and enterprise logic, should nonetheless be confirmed by the human in
the loop.

For a legacy modernization challenge involving a migration from AngularJS
to React, I’d charge this expertise an absolute 10/10. This functionality
opens up the likelihood for any people with respectable technical
experience and robust area data emigrate any legacy codebase to a
fashionable stack with minimal effort and in considerably much less time.

I consider that with a bottom-up strategy, breaking the issue down
into atomic elements, and clearly defining finest practices and
tips, AI-generated code might significantly speed up supply
timelines—even for complicated brownfield initiatives as we noticed for Bahmni.

The preliminary evaluation and the following assessment by specialists leads to a
crisp sufficient doc that lets us use the restricted house within the context
window in an environment friendly method so we will match extra data into one single
immediate. Successfully, this permits the LLM to investigate code in a method that’s
not restricted by how the code is organized within the first place by builders.
This additionally leads to decreasing the general value of utilizing LLMs, as a brute
drive strategy would imply that you just spend 10 occasions as a lot even for a a lot
less complicated challenge.

Whereas modernizing the legacy codebase is the primary product of this
proposed strategy, it isn’t the one beneficial one. The documentation
generated concerning the system is efficacious when supplied not simply to the top
customers / implementers in complementing or filling gaps in present techniques
documentation and in addition would stand in as a data base concerning the system
for ahead engineering groups pairing with LLMs to reinforce or enrich
system capabilities.

Why the Evaluation Section Issues

A key enabler of this profitable migration was a well-structured plan
and detailed scope assessment part previous to implementation. This early
funding paid dividends through the code era part. And not using a
clear understanding of the information stream, configuration construction, and
show logic, the AI would have struggled to supply coherent and
maintainable outputs. When you have labored with AI earlier than, you could have
seen that it’s persistently desperate to generate output. In an earlier
try, I proceeded with out enough warning and skipped the assessment
step—solely to find that the generated code included a useMemo hook
for an operation that was computationally trivial. One of many success
standards within the generated evaluation was that the code needs to be
performant, and this gave the impression to be the AI’s method of fulfilling that
requirement.

Apparently, the AI even added unit assessments to validate the
efficiency of that particular operation. Nevertheless, none of this was
explicitly required. It arose solely as a consequence of a poorly outlined intent. AI
integrated these modifications with out hesitation, regardless of not absolutely
understanding the underlying necessities or searching for clarification.
Reviewing each the generated evaluation and the corresponding code ensures
that unintended additions are recognized early and that deviations from
the unique expectations are minimized.

Evaluation additionally performs a key position in avoiding pointless back-and-forth
with the AI through the rebuild part. As an illustration, whereas refining the
immediate for the “Show Management Implementation
Information”,
I initially didn’t have the part specifying the unit assessments to be
included. Because of this, the AI generated a check that was largely
meaningless—providing a false sense of check protection with no actual
connection to the code below check.

Determine 7: AI generated unit check that verifies actuality
remains to be actual

In an try to repair this check, I started prompting
extensively—offering examples and detailed directions on how the unit
check needs to be structured. Nevertheless, the extra I prompted, the additional the
course of deviated from the unique goal of rebuilding the show
management. The main focus shifted solely to resolving unit check points, with
the AI even starting to assessment unrelated assessments within the codebase and
suggesting fixes for issues it recognized there.

Finally, realizing the rising divergence from the meant
job, I restarted the method with clearly outlined directions from the
outset, which proved to be far more practical.

This leads us to an important perception: Do not Interrupt AI.

LLMs, at their core, are predictive sequence mills that construct
narratives token by token. If you interrupt a mannequin mid-stream to
course-correct, you break the logical stream it was setting up.
Stanford’s “Misplaced within the
Center”
research revealed that fashions can undergo as much as a 20%
drop in accuracy when essential data is buried in the midst of
lengthy contexts, versus when it’s clearly framed upfront. This underscores
why beginning with a well-defined immediate and letting the AI full its
job unimpeded usually yields higher outcomes than fixed backtracking or
mid-flight corrections.

This concept can also be bolstered in “Why Human Intent Issues Extra as AI
Capabilities Develop” by Nick
Baumann,
which argues that as mannequin capabilities scale, clear human intent—not
simply brute mannequin power—turns into the important thing to unlocking helpful output.
Moderately than micromanaging each response, practitioners profit most by
designing clear, unambiguous setups and letting the AI full the arc
with out interruption.

Conclusion

You will need to make clear that this strategy isn’t meant to be a
silver bullet able to executing a large-scale migration with out
oversight. Moderately, its power lies in its potential to considerably
cut back improvement time—probably by a number of weeks—whereas sustaining
high quality and management.

The objective is not to interchange human experience however to amplify it—to
speed up supply timelines whereas making certain that high quality and
maintainability are preserved, if not improved, through the transition.

Additionally it is essential to notice that the expertise and outcomes mentioned
up to now are restricted to read-only controls. Extra complicated or interactive
elements could current extra challenges that require additional
analysis and refinement of the prompts used.

One of many key insights from exploring GenAI for legacy migration is
that whereas massive language fashions (LLMs) excel at general-purpose duties and
predefined workflows, their true potential in large-scale enterprise
transformation is just realized when guided by human experience. That is
nicely illustrated by Moravec’s Paradox, which observes that duties perceived
as intellectually complicated—reminiscent of logical reasoning—are comparatively simpler
for AI, whereas duties requiring human instinct and contextual
understanding stay difficult. Within the context of legacy modernization,
this paradox reinforces the significance of material specialists (SMEs)
and area specialists, whose deep expertise, contextual understanding,
and instinct are indispensable. Their experience allows extra correct
interpretation of necessities, validation of AI-generated outputs, and
knowledgeable decision-making—in the end making certain that the transformation is
aligned with the group’s objectives and constraints.

Whereas project-specific complexities could render this strategy formidable,
I consider that by adopting this structured workflow, AI-generated code can
considerably speed up supply timelines—even within the context of complicated
brownfield initiatives. The intent is to not exchange human experience, however to
increase it—streamlining improvement whereas safeguarding, and probably
enhancing, code high quality and maintainability. Though the standard and
architectural soundness of the legacy system stay essential elements, this
methodology gives a powerful start line. It reduces handbook overhead,
creates ahead momentum, and lays the groundwork for cleaner and extra
maintainable implementations by way of expert-led, guided refactoring.

I firmly consider following this workflow opens up the likelihood for
any people with respectable technical experience and robust area
data emigrate any legacy codebase to a contemporary stack with minimal
effort and in considerably much less time.